CN107657249A - Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again - Google Patents
Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again Download PDFInfo
- Publication number
- CN107657249A CN107657249A CN201711017356.6A CN201711017356A CN107657249A CN 107657249 A CN107657249 A CN 107657249A CN 201711017356 A CN201711017356 A CN 201711017356A CN 107657249 A CN107657249 A CN 107657249A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- neural network
- network model
- model
- identified
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses method, apparatus, storage medium and the processor that a kind of Analysis On Multi-scale Features pedestrian identifies again.Wherein, this method includes：First convolution neural network model is established according to default Multi resolution feature extraction method；Neural network model is accumulated according to the default public data collection training first volume, obtains the second convolution neural network model；The model group being made up of the second convolution neural network model is trained according to default ternary group data set, obtains the 3rd convolutional neural networks model, wherein, model group includes three the second convolution neural network models in parallel；Target pedestrian image and pedestrian image to be identified are inputted to the 3rd convolutional neural networks model, obtains target pedestrian characteristic vector and pedestrian's characteristic vector to be identified；The vector distance between target pedestrian characteristic vector and pedestrian's characteristic vector to be identified is calculated, result is identified according to vector distance.The present invention solves the relatively low technical problem of accuracy of identification existing for pedestrian's weight identification method of the prior art.
Description
Technical field
The present invention relates to field of video image processing, the side identified again in particular to a kind of Analysis On Multi-scale Features pedestrian
Method, device, storage medium and processor.
Background technology
Pedestrian's weight identification technology, i.e., at multiple visual angles without automatically retrieval target pedestrian between the camera video of juxtaposition
Technology, for intelligent video monitoring, suspicion personnel retrieval etc. application play vital effect.However, due to camera
Visual angle difference, pedestrian's attitudes vibration, detection error early stage etc. influence, and pedestrian identifies problem again, and there is very big challenge.
Existing pedestrian's weight identification technology, by extracting the global and local feature of pedestrian, alleviate due to attitudes vibration, inspection
Influence caused by surveying error etc..Patent CN104376334A proposes a kind of pedestrian's comparison method of multi-scale feature fusion, leads to
The scaling that different proportion is carried out to image is crossed, low yardstick and high yardstick are combined, are compared on different images yardstick.So
And this method uses the feature of hand-designed, poor robustness, and two kinds of graphical rules are only included, precision is low, and the scope of application is small.
Document《Person Re-Identification by Deep Joint Learning of Multi-Loss
Classification》A kind of convolutional neural networks of multiple-limb are proposed, global characteristics are extracted based on complete characterization figure, and will
Sharing feature figure horizontal segmentation is in strip sub-block and extracts local feature respectively, realizes that pedestrian identifies again jointly.However, sub-block number
More, the detail of local feature is stronger, and the complexity of algorithm is also higher, and its precision is influenceed by detection error early stage, difficult
To be guaranteed.Document《Pose Invariant Embedding for Deep Person Re-Identification》Knot
Attitude estimation technology is closed, based on the different human body position detected, carries out feature extraction and the similarity mode of corresponding position.Should
Method effectively can identify reliable again using the similarity between the local feature of different pedestrian image corresponding positions, lifting pedestrian
Property, but the pedestrian detection and location detection to early stage require higher, and the error band of location detection carrys out the accumulation of error, influences to know
Other precision.To sum up, there is the relatively low technical problem of accuracy of identification in pedestrian's weight identification method of the prior art.
For it is above-mentioned the problem of, not yet propose effective solution at present.
The content of the invention
The embodiments of the invention provide method, apparatus, storage medium and the processing that a kind of Analysis On Multi-scale Features pedestrian identifies again
Device, at least to solve the relatively low technical problem of accuracy of identification existing for pedestrian's weight identification method of the prior art.
One side according to embodiments of the present invention, there is provided a kind of Analysis On Multi-scale Features pedestrian knows method for distinguishing, the party again
Method includes：First convolution neural network model is established according to default Multi resolution feature extraction method, wherein, above-mentioned first convolutional Neural
The input branch of random layer includes the output branch of at least one anteposition level of above-mentioned random layer in network model；According to
The above-mentioned first convolution neural network model of default public data collection training, obtains the second convolution neural network model, wherein, it is above-mentioned
Second convolution neural network model is the above-mentioned first convolution neural network model for reaching convergence state；According to default triple number
The model group being made up of according to set pair above-mentioned second convolution neural network model is trained, and obtains the 3rd convolutional neural networks mould
Type, wherein, above-mentioned model group includes three above-mentioned second convolution neural network models in parallel, above-mentioned 3rd convolutional Neural net
Network model is the above-mentioned model group for reaching convergence state；Target pedestrian image and pedestrian image to be identified are inputted to the above-mentioned 3rd
Convolutional neural networks model, obtain target pedestrian characteristic vector and pedestrian's characteristic vector to be identified；It is special to calculate above-mentioned target pedestrian
The vector distance between the above-mentioned pedestrian's characteristic vector to be identified of vector sum is levied, result is identified according to above-mentioned vector distance.
Further, according to default ternary group data set to the model that is made up of above-mentioned second convolution neural network model
Before group is trained, the above method also includes：Obtain pedestrian's video image that multiple cameras photograph；Intercept above-mentioned row
The pedestrian area in every frame picture in people's video image, and identity label is added to above-mentioned pedestrian area；According to above-mentioned pedestrian
Region and above-mentioned identity label obtain pedestrian's weight identification data collection；Data combination is carried out to above-mentioned pedestrian weight identification data collection, obtained
To above-mentioned default ternary group data set.
Further, above-mentioned basis presets ternary group data set to the mould that is made up of above-mentioned second convolution neural network model
Type group is trained, and obtaining the 3rd convolutional neural networks model includes：Will be any one in above-mentioned default ternary group data set
Group triple training picture is separately input into above three above-mentioned second convolution neural network model in parallel, so as to obtain three
Individual full articulamentum characteristic vector, wherein, the full articulamentum characteristic vector of above three above-mentioned second convolution in parallel with above three
Neural network model has one-to-one relationship；Calculated based on the full articulamentum characteristic vector of above three and trained with above-mentioned triple
The cost function of picture association；According to above-mentioned cost function and default the Stochastic gradient method above-mentioned volume Two in parallel to above three
The weights of product neural network model synchronize renewal, obtain above-mentioned 3rd convolutional neural networks model.
Further, between the above-mentioned target pedestrian characteristic vector of above-mentioned calculating and above-mentioned pedestrian's characteristic vector to be identified to
Span is from being identified result according to above-mentioned vector distance includes：Judge whether above-mentioned vector distance is less than pre-determined distance threshold value；
If above-mentioned vector distance is less than above-mentioned pre-determined distance threshold value, it is determined that above-mentioned target pedestrian image and above-mentioned pedestrian image to be identified
Matching；If above-mentioned vector distance is not less than above-mentioned pre-determined distance threshold value, it is determined that above-mentioned target pedestrian image and above-mentioned to be identified
Pedestrian image mismatches.
Another aspect according to embodiments of the present invention, the device that a kind of Analysis On Multi-scale Features pedestrian identifies again is additionally provided, should
Device includes：First construction unit, for establishing the first convolution neural network model according to default Multi resolution feature extraction method, its
In, the input branch of random layer includes at least one anteposition of above-mentioned random layer in above-mentioned first convolution neural network model
The output branch of level；First training unit, for according to the above-mentioned first convolutional neural networks mould of default public data collection training
Type, the second convolution neural network model is obtained, wherein, above-mentioned second convolution neural network model is reach convergence state above-mentioned
First convolution neural network model；Second training unit, for the default ternary group data set of basis to by above-mentioned second convolution god
The model group formed through network model is trained, and obtains the 3rd convolutional neural networks model, wherein, above-mentioned model group bag
Containing three above-mentioned second convolution neural network models in parallel, above-mentioned 3rd convolutional neural networks model is to reach convergence state
Above-mentioned model group；Input block, for inputting target pedestrian image and pedestrian image to be identified to above-mentioned 3rd convolutional Neural
Network model, obtain target pedestrian characteristic vector and pedestrian's characteristic vector to be identified；Computing unit, for calculating above-mentioned target line
Vector distance between people's characteristic vector and above-mentioned pedestrian's characteristic vector to be identified, is identified tying according to above-mentioned vector distance
Fruit.
Further, said apparatus also includes：First acquisition unit, regarded for obtaining the pedestrian that multiple cameras photograph
Frequency image；First processing units, for intercepting the pedestrian area in every frame picture in above-mentioned pedestrian's video image, and to above-mentioned
Pedestrian area adds identity label；Second acquisition unit, for obtaining pedestrian according to above-mentioned pedestrian area and above-mentioned identity label
Weight identification data collection；Second processing unit, for carrying out data combination to above-mentioned pedestrian weight identification data collection, obtain above-mentioned default
Ternary group data set.
Further, above-mentioned second training unit includes：Subelement is inputted, for by above-mentioned default ternary group data set
Any one group of triple training picture be separately input into above three above-mentioned second convolution neural network model in parallel, from
And three full articulamentum characteristic vectors are obtained, wherein, above-mentioned in parallel with above three of the full articulamentum characteristic vector of above three
Second convolution neural network model has one-to-one relationship；Computation subunit, for based on the full articulamentum feature of above three
Vector calculates the cost function associated with above-mentioned triple training picture；Update subelement, for according to above-mentioned cost function and
The weights of the default Stochastic gradient method above-mentioned second convolution neural network model in parallel to above three synchronize renewal, obtain
Above-mentioned 3rd convolutional neural networks model.
Further, above-mentioned computing unit includes：Judgment sub-unit, for judging it is default whether above-mentioned vector distance is less than
Distance threshold；First determination subelement, if being less than above-mentioned pre-determined distance threshold value for above-mentioned vector distance, it is determined that above-mentioned target
Pedestrian image and above-mentioned pedestrian image matching to be identified；Second determination subelement, if for above-mentioned vector distance not less than above-mentioned
Pre-determined distance threshold value, it is determined that above-mentioned target pedestrian image and above-mentioned pedestrian image to be identified mismatch.
Another aspect according to embodiments of the present invention, additionally provides a kind of storage medium, and above-mentioned storage medium includes storage
Program, wherein, equipment where above-mentioned storage medium is controlled when said procedure is run performs above-mentioned Analysis On Multi-scale Features pedestrian
Method for distinguishing is known again.
Another aspect according to embodiments of the present invention, additionally provides a kind of processor, and above-mentioned processor is used for operation program,
Wherein, perform above-mentioned Analysis On Multi-scale Features pedestrian when said procedure is run and know method for distinguishing again.
In embodiments of the present invention, the side of the first convolution neural network model is established using default Multi resolution feature extraction method
Formula, wherein, the input branch of random layer includes at least one anteposition layer of random layer in the first convolution neural network model
The output branch of level；Neural network model is accumulated by the default public data collection training first volume, obtains the second convolutional neural networks
Model, wherein, the second convolution neural network model is the first convolution neural network model for reaching convergence state；According to default three
The model group that tuple data set pair is made up of the second convolution neural network model is trained, and obtains the 3rd convolutional neural networks
Model, wherein, model group includes three the second convolution neural network models in parallel, the 3rd convolutional neural networks model be up to
To the model group of convergence state；Target pedestrian image and pedestrian image to be identified are inputted to the 3rd convolutional neural networks model,
Obtain target pedestrian characteristic vector and pedestrian's characteristic vector to be identified；Calculating target pedestrian characteristic vector and row to be identified are reached
Vector distance between people's characteristic vector, the purpose for being identified according to vector distance result, it is achieved thereby that lifting pedestrian's weight
The technique effect of the accuracy of identification of identification method, the recognition efficiency of raising pedestrian's weight identification method, and then solve prior art
In pedestrian's weight identification method existing for the relatively low technical problem of accuracy of identification.
Brief description of the drawings
Accompanying drawing described herein is used for providing a further understanding of the present invention, forms the part of the application, this hair
Bright schematic description and description is used to explain the present invention, does not form inappropriate limitation of the present invention.In the accompanying drawings：
Fig. 1 is the flow signal that a kind of optional Analysis On Multi-scale Features pedestrian according to embodiments of the present invention knows method for distinguishing again
Figure；
Fig. 2 is that the flow that the optional Analysis On Multi-scale Features pedestrian of another kind according to embodiments of the present invention knows method for distinguishing again is shown
It is intended to；
Fig. 3 is that the flow that another optional Analysis On Multi-scale Features pedestrian according to embodiments of the present invention knows method for distinguishing again is shown
It is intended to；
Fig. 4 is that the flow that another optional Analysis On Multi-scale Features pedestrian according to embodiments of the present invention knows method for distinguishing again is shown
It is intended to；
Fig. 5 is the structural representation for the device that a kind of optional Analysis On Multi-scale Features pedestrian according to embodiments of the present invention identifies again
Figure；
Fig. 6 is a kind of structural representation of optional first convolution neural network model according to embodiments of the present invention.
Embodiment
In order that those skilled in the art more fully understand the present invention program, below in conjunction with the embodiment of the present invention
Accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is only
The embodiment of a part of the invention, rather than whole embodiments.Based on the embodiment in the present invention, ordinary skill people
The every other embodiment that member is obtained under the premise of creative work is not made, it should all belong to the model that the present invention protects
Enclose.
It should be noted that term " first " in description and claims of this specification and above-mentioned accompanying drawing, "
Two " etc. be for distinguishing similar object, without for describing specific order or precedence.It should be appreciated that so use
Data can exchange in the appropriate case, so as to embodiments of the invention described herein can with except illustrating herein or
Order beyond those of description is implemented.In addition, term " comprising " and " having " and their any deformation, it is intended that cover
Cover it is non-exclusive include, be not necessarily limited to for example, containing the process of series of steps or unit, method, system, product or equipment
Those steps or unit clearly listed, but may include not list clearly or for these processes, method, product
Or the intrinsic other steps of equipment or unit.
Embodiment 1
According to embodiments of the present invention, there is provided a kind of Analysis On Multi-scale Features pedestrian knows the embodiment of method for distinguishing, it is necessary to say again
It is bright, it can be held the step of the flow of accompanying drawing illustrates in the computer system of such as one group computer executable instructions
OK, although also, show logical order in flow charts, in some cases, can be with different from order herein
Perform shown or described step.
Fig. 1 is the flow signal that a kind of optional Analysis On Multi-scale Features pedestrian according to embodiments of the present invention knows method for distinguishing again
Figure, as shown in figure 1, this method comprises the following steps：
Step S102, the first convolution neural network model is established according to default Multi resolution feature extraction method, wherein, the first volume
The input branch of random layer includes the output branch of at least one anteposition level of random layer in product neural network model；
Step S104, neural network model is accumulated according to the default public data collection training first volume, obtains the second convolutional Neural
Network model, wherein, the second convolution neural network model is the first convolution neural network model for reaching convergence state；
Step S106, the model group being made up of the second convolution neural network model is entered according to default ternary group data set
Row training, obtains the 3rd convolutional neural networks model, wherein, model group includes three the second convolutional neural networks moulds in parallel
Type, the 3rd convolutional neural networks model are the model group for reaching convergence state；
Step S108, target pedestrian image and pedestrian image to be identified are inputted to the 3rd convolutional neural networks model, is obtained
Target pedestrian characteristic vector and pedestrian's characteristic vector to be identified；
Step S110, the vector distance between target pedestrian characteristic vector and pedestrian's characteristic vector to be identified is calculated, according to
Vector distance is identified result.
In embodiments of the present invention, the side of the first convolution neural network model is established using default Multi resolution feature extraction method
Formula, wherein, the input branch of random layer includes at least one anteposition layer of random layer in the first convolution neural network model
The output branch of level；Neural network model is accumulated by the default public data collection training first volume, obtains the second convolutional neural networks
Model, wherein, the second convolution neural network model is the first convolution neural network model for reaching convergence state；According to default three
The model group that tuple data set pair is made up of the second convolution neural network model is trained, and obtains the 3rd convolutional neural networks
Model, wherein, model group includes three the second convolution neural network models in parallel, the 3rd convolutional neural networks model be up to
To the model group of convergence state；Target pedestrian image and pedestrian image to be identified are inputted to the 3rd convolutional neural networks model,
Obtain target pedestrian characteristic vector and pedestrian's characteristic vector to be identified；Calculating target pedestrian characteristic vector and row to be identified are reached
Vector distance between people's characteristic vector, the purpose for being identified according to vector distance result, it is achieved thereby that lifting pedestrian's weight
The technique effect of the accuracy of identification of identification method, the recognition efficiency of raising pedestrian's weight identification method, and then solve prior art
In pedestrian's weight identification method existing for the relatively low technical problem of accuracy of identification.
Alternatively, in step S102, the input of each level in the first convolution neural network model can include multiple
The output of level, for extracting the feature of different scale.For example, the input of the n-th layer level of the first convolutional network can include the
(N-1) output, the output of (N-3) level and the output of (N-5) level of level.
Alternatively, the first convolution neural network model in step S102 includes multiple multiple dimensioned extraction units, Mei Gedan
The input of convolutional layer in member can be the above hierarchical output of all levels or portion, therefore, can reuse forefront layer
The feature of level extraction, be advantageous to gradient passback during training, lift the convergence rate of network.In addition, this model structure can carry
The characteristic vector of a variety of yardsticks is taken, overall situation and partial situation's information of pedestrian image is effectively utilized, hence for conditions such as posture, visual angles
Change keeps robustness, lifting pedestrian's weight accuracy of identification.
Alternatively, in step S104, public data collection can be object classification data set, such as ImageNet.According to pre-
If public data collection training first volume product neural network model can use classification cross entropy cost function.
Alternatively, in step S106, according to default ternary group data set to being made up of the second convolution neural network model
Model group is trained, it is ensured that and the distance between same pedestrian image is less than the distance between different pedestrian images,
So as to which the feature learnt has good distribution, ensure accuracy of identification.
Alternatively, in step S110, the vector between target pedestrian characteristic vector and pedestrian's characteristic vector to be identified is calculated
Distance, being identified result according to vector distance includes：Bring the vector distance into preset matching rule, be identified tying
Fruit, the preset matching rule can be the distance between target pedestrian characteristic vector and pedestrian's characteristic vector to be identified less than default
Distance threshold or the distance between multiple pedestrian's characteristic vectors to be identified and target pedestrian's characteristic vector in sort most
Small preceding M result, for example, M can value be 5.
Alternatively, Fig. 2 is that the optional Analysis On Multi-scale Features pedestrian of another kind according to embodiments of the present invention knows method for distinguishing again
Schematic flow sheet, as shown in Fig. 2 before step S106 is performed, i.e., according to default ternary group data set to by volume Two
Before the model group of product neural network model composition is trained, this method can also include：
Step S202, obtain pedestrian's video image that multiple cameras photograph；
Step S204, the pedestrian area in every frame picture in pedestrian's video image is intercepted, and body is added to pedestrian area
Part label；
Step S206, pedestrian's weight identification data collection is obtained according to pedestrian area and identity label；
Step S208, data combination is carried out to pedestrian's weight identification data collection, obtains default ternary group data set.
Alternatively, step S204 is performed to specifically include：Pedestrian's video image is decoded, obtains multiple single frames pedestrian figure
Picture, to the pedestrian area point of addition label and identity label in every frame pedestrian image；
Alternatively, step S208 is performed to specifically include：Data combination is carried out to pedestrian's weight identification data collection, obtains default three
Tuple data collection, wherein, each triple in the default ternary group data set includes two the image from same pedestrian
And the image of other pedestrians, the mode of data combination can be the merging of data acquisition system.
Alternatively, Fig. 3 is that the optional Analysis On Multi-scale Features pedestrian of another according to embodiments of the present invention knows method for distinguishing again
Schematic flow sheet, as shown in figure 3, perform step S106, i.e., according to preset ternary group data set to by the second convolution nerve net
The model group of network model composition is trained, and obtaining the 3rd convolutional neural networks model includes：
Step S302, any one group of triple training picture in default ternary group data set is separately input into three simultaneously
In second convolution neural network model of connection, so as to obtain three full articulamentum characteristic vectors, wherein, three full articulamentum features
The vector second convolution neural network model in parallel with three has one-to-one relationship；
Step S304, the cost function associated with triple training picture is calculated based on three full articulamentum characteristic vectors；
Step S306, according to cost function and default the Stochastic gradient method second convolution neural network model in parallel to three
Weights synchronize renewal, obtain the 3rd convolutional neural networks model.
Alternatively, the function formula of the cost function in step S306 can be：E=[d (x, x_p)-d (x, x_n)+τ],
Wherein, d is distance metric function, and x is the probe images in triple, and x_p is positive example image, and x_n be counter-example image, τ for away from
From controlling elements, [] if+represent in bracket value as negative, E 0, if positive number, then E is equal to the value.Alternatively, apart from degree
Flow function d can be Euclidean distance.
Alternatively, Fig. 4 is that the optional Analysis On Multi-scale Features pedestrian of another according to embodiments of the present invention knows method for distinguishing again
Schematic flow sheet, as shown in figure 4, perform step S106, that is, calculate target pedestrian characteristic vector and pedestrian's feature to be identified to
Vector distance between amount, being identified result according to vector distance includes：
Step S402, judges whether vector distance is less than pre-determined distance threshold value；
Step S404, if vector distance is less than pre-determined distance threshold value, it is determined that target pedestrian image and pedestrian to be identified figure
As matching；
Step S406, if vector distance is not less than pre-determined distance threshold value, it is determined that target pedestrian image and pedestrian to be identified
Image mismatches.
In embodiments of the present invention, the side of the first convolution neural network model is established using default Multi resolution feature extraction method
Formula, wherein, the input branch of random layer includes at least one anteposition layer of random layer in the first convolution neural network model
The output branch of level；Neural network model is accumulated by the default public data collection training first volume, obtains the second convolutional neural networks
Model, wherein, the second convolution neural network model is the first convolution neural network model for reaching convergence state；According to default three
The model group that tuple data set pair is made up of the second convolution neural network model is trained, and obtains the 3rd convolutional neural networks
Model, wherein, model group includes three the second convolution neural network models in parallel, the 3rd convolutional neural networks model be up to
To the model group of convergence state；Target pedestrian image and pedestrian image to be identified are inputted to the 3rd convolutional neural networks model,
Obtain target pedestrian characteristic vector and pedestrian's characteristic vector to be identified；Calculating target pedestrian characteristic vector and row to be identified are reached
Vector distance between people's characteristic vector, the purpose for being identified according to vector distance result, it is achieved thereby that lifting pedestrian's weight
The technique effect of the accuracy of identification of identification method, the recognition efficiency of raising pedestrian's weight identification method, and then solve prior art
In pedestrian's weight identification method existing for the relatively low technical problem of accuracy of identification.
Embodiment 2
Another aspect according to embodiments of the present invention, the device that a kind of Analysis On Multi-scale Features pedestrian identifies again is additionally provided, such as
Shown in Fig. 5, the device includes：
First construction unit 501, for establishing the first convolution neural network model according to default Multi resolution feature extraction method,
Wherein, the input branch of random layer includes at least one anteposition level of random layer in the first convolution neural network model
Output branch；First training unit 503, for according to default public data collection training first volume product neural network model, obtaining
Second convolution neural network model, wherein, the second convolution neural network model is the first convolution nerve net for reaching convergence state
Network model；Second training unit 505, for according to default ternary group data set to being made up of the second convolution neural network model
Model group is trained, and obtains the 3rd convolutional neural networks model, wherein, model group includes three the second convolution in parallel
Neural network model, the 3rd convolutional neural networks model are the model group for reaching convergence state；Input block 507, for defeated
Enter target pedestrian image and pedestrian image to be identified to the 3rd convolutional neural networks model, obtain target pedestrian characteristic vector and treat
Identify pedestrian's characteristic vector；Computing unit 509, for calculating between target pedestrian characteristic vector and pedestrian's characteristic vector to be identified
Vector distance, result is identified according to vector distance.
Alternatively, Fig. 6 is that a kind of structure of optional first convolution neural network model according to embodiments of the present invention is shown
It is intended to, as shown in fig. 6, the first convolution neural network model includes：
First convolutional layer, convolution kernel size are 5x5, and port number 16, step-length 2, activation primitive type is ReLU.
First pond layer, pond size are 2x2, step-length 2.
Second convolutional layer, convolution kernel size are 3x3, and port number 16, step-length 2, activation primitive type is ReLU.
Second pond layer, pond size are 2x2, step-length 2.
First multiple dimensioned extraction unit, comprising 3 convolutional layers, every layer of convolution kernel size is 3x3, port number 12, step-length
For 1, activation primitive type is ReLU, is normalized using batch normalization.The input of every layer of convolutional layer is single
Above the output characteristic figure of all convolutional layers splices obtained total characteristic figure in first.
First Transition layer, is realized by convolution, and convolution kernel size is 3x3, port number 128, step-length 2, activation primitive class
Type is ReLU.
Second multiple dimensioned extraction unit, comprising 3 convolutional layers, every layer of convolution kernel size is 3x3, port number 12, step-length
For 1, activation primitive type is ReLU, is normalized using batch normalization.The input of every layer of convolutional layer is single
Above the output characteristic figure of all convolutional layers splices obtained total characteristic figure in first.
Second transition zone, is realized by convolution, and convolution kernel size is 3x3, port number 256, step-length 2, activation primitive class
Type is ReLU.
3rd multiple dimensioned extraction unit, comprising 3 convolutional layers, every layer of convolution kernel size is 3x3, port number 12, step-length
For 1, activation primitive type is ReLU, is normalized using batch normalization.The input of every layer of convolutional layer is single
Above the output characteristic figure of all convolutional layers splices obtained total characteristic figure in first.
3rd transition zone, is realized by convolution, and convolution kernel size is 3x3, port number 512, step-length 2, activation primitive class
Type is ReLU.
4th multiple dimensioned extraction unit, comprising 3 convolutional layers, every layer of convolution kernel size is 3x3, port number 12, step-length
For 1, activation primitive type is ReLU, is normalized using batch normalization.The input of every layer of convolutional layer is single
Above the output characteristic figure of all convolutional layers splices obtained total characteristic figure in first.
The average pond layer of the overall situation, the output characteristic figure inputted as the 4th multiple dimensioned all convolutional layers of extraction unit splice to obtain
Total characteristic figure by batch normalization normalization after result.
Full articulamentum, include 512 neurons.
Output layer, using softmax classification functions, neuron number is classification number.For ImageNet data sets, nerve
First number is 1000.
Alternatively, the device can also include：First acquisition unit, regarded for obtaining the pedestrian that multiple cameras photograph
Frequency image；First processing units, for intercepting the pedestrian area in every frame picture in pedestrian's video image, and to pedestrian area
Add identity label；Second acquisition unit, for obtaining pedestrian's weight identification data collection according to pedestrian area and identity label；Second
Processing unit, for carrying out data combination to pedestrian's weight identification data collection, obtain default ternary group data set.
Alternatively, the second training unit can include：Subelement is inputted, for will be any in default ternary group data set
One group of triple training picture is separately input into three the second convolution neural network models in parallel, is connected entirely so as to obtain three
A layer characteristic vector is connect, wherein, three full articulamentum characteristic vectors, the second convolution neural network model in parallel with three has one
One corresponding relation；Computation subunit, for calculating what is associated with triple training picture based on three full articulamentum characteristic vectors
Cost function；Subelement is updated, for according to cost function and default Stochastic gradient method second convolutional Neural in parallel to three
The weights of network model synchronize renewal, obtain the 3rd convolutional neural networks model.
Alternatively, computing unit includes：Judgment sub-unit, for judging whether vector distance is less than pre-determined distance threshold value；
First determination subelement, if being less than pre-determined distance threshold value for vector distance, it is determined that target pedestrian image and pedestrian to be identified
Images match；Second determination subelement, if being not less than pre-determined distance threshold value for vector distance, it is determined that target pedestrian image and
Pedestrian image to be identified mismatches.
Another aspect according to embodiments of the present invention, additionally provides a kind of storage medium, and the storage medium includes storage
Program, wherein, control equipment where storage medium to perform the Analysis On Multi-scale Features pedestrian weight in above-described embodiment 1 when program is run
Know method for distinguishing.
Another aspect according to embodiments of the present invention, additionally provides a kind of processor, and the processor is used for operation program, its
In, the Analysis On Multi-scale Features pedestrian that program is performed in above-described embodiment 1 when running knows method for distinguishing again.
In embodiments of the present invention, the side of the first convolution neural network model is established using default Multi resolution feature extraction method
Formula, wherein, the input branch of random layer includes at least one anteposition layer of random layer in the first convolution neural network model
The output branch of level；Neural network model is accumulated by the default public data collection training first volume, obtains the second convolutional neural networks
Model, wherein, the second convolution neural network model is the first convolution neural network model for reaching convergence state；According to default three
The model group that tuple data set pair is made up of the second convolution neural network model is trained, and obtains the 3rd convolutional neural networks
Model, wherein, model group includes three the second convolution neural network models in parallel, the 3rd convolutional neural networks model be up to
To the model group of convergence state；Target pedestrian image and pedestrian image to be identified are inputted to the 3rd convolutional neural networks model,
Obtain target pedestrian characteristic vector and pedestrian's characteristic vector to be identified；Calculating target pedestrian characteristic vector and row to be identified are reached
Vector distance between people's characteristic vector, the purpose for being identified according to vector distance result, it is achieved thereby that lifting pedestrian's weight
The technique effect of the accuracy of identification of identification method, the recognition efficiency of raising pedestrian's weight identification method, and then solve prior art
In pedestrian's weight identification method existing for the relatively low technical problem of accuracy of identification.
The embodiments of the present invention are for illustration only, do not represent the quality of embodiment.
In the above embodiment of the present invention, the description to each embodiment all emphasizes particularly on different fields, and does not have in some embodiment
The part of detailed description, it may refer to the associated description of other embodiment.
In several embodiments provided herein, it should be understood that disclosed technology contents, others can be passed through
Mode is realized.Wherein, device embodiment described above is only schematical, such as the division of the unit, Ke Yiwei
A kind of division of logic function, can there is an other dividing mode when actually realizing, for example, multiple units or component can combine or
Person is desirably integrated into another system, or some features can be ignored, or does not perform.Another, shown or discussed is mutual
Between coupling or direct-coupling or communication connection can be INDIRECT COUPLING or communication link by some interfaces, unit or module
Connect, can be electrical or other forms.
The unit illustrated as separating component can be or may not be physically separate, show as unit
The part shown can be or may not be physical location, you can with positioned at a place, or can also be distributed to multiple
On unit.Some or all of unit therein can be selected to realize the purpose of this embodiment scheme according to the actual needs.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, can also
That unit is individually physically present, can also two or more units it is integrated in a unit.Above-mentioned integrated list
Member can both be realized in the form of hardware, can also be realized in the form of SFU software functional unit.
If the integrated unit is realized in the form of SFU software functional unit and is used as independent production marketing or use
When, it can be stored in a computer read/write memory medium.Based on such understanding, technical scheme is substantially
The part to be contributed in other words to prior art or all or part of the technical scheme can be in the form of software products
Embody, the computer software product is stored in a storage medium, including some instructions are causing a computer
Equipment (can be personal computer, server or network equipment etc.) perform each embodiment methods described of the present invention whole or
Part steps.And foregoing storage medium includes：USB flash disk, read-only storage (ROM, Read-Only Memory), arbitrary access are deposited
Reservoir (RAM, Random Access Memory), mobile hard disk, magnetic disc or CD etc. are various can be with store program codes
Medium.
Described above is only the preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications also should
It is considered as protection scope of the present invention.
Claims (10)
1. a kind of Analysis On Multi-scale Features pedestrian knows method for distinguishing again, it is characterised in that including：
First convolution neural network model is established according to default Multi resolution feature extraction method, wherein, the first convolution nerve net
The input branch of random layer includes the output branch of at least one anteposition level of the random layer in network model；
According to default public data collection training the first convolution neural network model, the second convolution neural network model is obtained,
Wherein, the second convolution neural network model is the first convolution neural network model for reaching convergence state；
The model group being made up of the second convolution neural network model is trained according to default ternary group data set, obtained
To the 3rd convolutional neural networks model, wherein, the model group includes three the second convolutional neural networks moulds in parallel
Type, the 3rd convolutional neural networks model are the model group for reaching convergence state；
Target pedestrian image and pedestrian image to be identified are inputted to the 3rd convolutional neural networks model, obtains target pedestrian spy
Levy vector sum pedestrian's characteristic vector to be identified；
Calculate the vector distance between the target pedestrian characteristic vector and pedestrian's characteristic vector to be identified, according to it is described to
Span is from being identified result.
2. according to the method for claim 1, it is characterised in that in the default ternary group data set of basis to by the volume Two
Before the model group of product neural network model composition is trained, methods described also includes：
Obtain pedestrian's video image that multiple cameras photograph；
The pedestrian area in every frame picture in pedestrian's video image is intercepted, and identity mark is added to the pedestrian area
Label；
Pedestrian's weight identification data collection is obtained according to the pedestrian area and the identity label；
Data combination is carried out to pedestrian weight identification data collection, obtains the default ternary group data set.
3. according to the method for claim 1, it is characterised in that the basis presets ternary group data set to by described second
The model group of convolutional neural networks model composition is trained, and obtaining the 3rd convolutional neural networks model includes：
By any one group of triple training picture in the default ternary group data set be separately input into described three it is in parallel
In the second convolution neural network model, so as to obtain three full articulamentum characteristic vectors, wherein, three full articulamentums
The characteristic vector second convolution neural network model in parallel with described three has one-to-one relationship；
The cost function associated with triple training picture is calculated based on described three full articulamentum characteristic vectors；
According to the cost function and default the Stochastic gradient method second convolution neural network model in parallel to described three
Weights synchronize renewal, obtain the 3rd convolutional neural networks model.
4. according to the method for claim 1, it is characterised in that described to calculate the target pedestrian characteristic vector and described treat
The vector distance between pedestrian's characteristic vector is identified, being identified result according to the vector distance includes：
Judge whether the vector distance is less than pre-determined distance threshold value；
If the vector distance is less than the pre-determined distance threshold value, it is determined that the target pedestrian image and the pedestrian to be identified
Images match；
If the vector distance is not less than the pre-determined distance threshold value, it is determined that the target pedestrian image and the row to be identified
People's image mismatches.
A kind of 5. device that Analysis On Multi-scale Features pedestrian identifies again, it is characterised in that including：
First construction unit, for establishing the first convolution neural network model according to default Multi resolution feature extraction method, wherein, institute
State at least one anteposition level that the input branch of random layer in the first convolution neural network model includes the random layer
Output branch；
First training unit, for according to default public data collection training the first convolution neural network model, obtaining second
Convolutional neural networks model, wherein, the second convolution neural network model is first convolution god for reaching convergence state
Through network model；
Second training unit, for according to default ternary group data set to the mould that is made up of the second convolution neural network model
Type group is trained, and obtains the 3rd convolutional neural networks model, wherein, the model group includes three in parallel described the
Two convolutional neural networks models, the 3rd convolutional neural networks model are the model group for reaching convergence state；
Input block, for inputting target pedestrian image and pedestrian image to be identified to the 3rd convolutional neural networks model,
Obtain target pedestrian characteristic vector and pedestrian's characteristic vector to be identified；
Computing unit, for calculate between the target pedestrian characteristic vector and pedestrian's characteristic vector to be identified to span
From being identified result according to the vector distance.
6. device according to claim 5, it is characterised in that described device also includes：
First acquisition unit, the pedestrian's video image photographed for obtaining multiple cameras；
First processing units, for intercepting the pedestrian area in every frame picture in pedestrian's video image, and to the row
Add identity label in people region；
Second acquisition unit, for obtaining pedestrian's weight identification data collection according to the pedestrian area and the identity label；
Second processing unit, for carrying out data combination to pedestrian weight identification data collection, obtain the default triple number
According to collection.
7. device according to claim 5, it is characterised in that second training unit includes：
Subelement is inputted, for any one group of triple training picture in the default ternary group data set to be separately input into
In described three the second convolution neural network models in parallel, so as to obtain three full articulamentum characteristic vectors, wherein, institute
Stating three full articulamentum characteristic vectors, the second convolution neural network model in parallel with described three, there is one-to-one corresponding to close
System；
Computation subunit, for calculating what is associated with triple training picture based on described three full articulamentum characteristic vectors
Cost function；
Subelement is updated, for according to the cost function and default the Stochastic gradient method volume Two in parallel to described three
The weights of product neural network model synchronize renewal, obtain the 3rd convolutional neural networks model.
8. device according to claim 5, it is characterised in that the computing unit includes：
Judgment sub-unit, for judging whether the vector distance is less than pre-determined distance threshold value；
First determination subelement, if being less than the pre-determined distance threshold value for the vector distance, it is determined that the target pedestrian
Image and the pedestrian image matching to be identified；
Second determination subelement, if being not less than the pre-determined distance threshold value for the vector distance, it is determined that the target line
People's image and the pedestrian image to be identified mismatch.
A kind of 9. storage medium, it is characterised in that the storage medium includes the program of storage, wherein, run in described program
When control the storage medium where equipment perform claim require 1 Analysis On Multi-scale Features into claim 4 described in any one
Pedestrian knows method for distinguishing again.
A kind of 10. processor, it is characterised in that the processor is used for operation program, wherein, right of execution when described program is run
Profit requires that the 1 Analysis On Multi-scale Features pedestrian into claim 4 described in any one knows method for distinguishing again.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711017356.6A CN107657249A (en) | 2017-10-26 | 2017-10-26 | Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711017356.6A CN107657249A (en) | 2017-10-26 | 2017-10-26 | Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107657249A true CN107657249A (en) | 2018-02-02 |
Family
ID=61094991
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711017356.6A Pending CN107657249A (en) | 2017-10-26 | 2017-10-26 | Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107657249A (en) |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108491880A (en) * | 2018-03-23 | 2018-09-04 | 西安电子科技大学 | Object classification based on neural network and position and orientation estimation method |
| CN108681743A (en) * | 2018-04-16 | 2018-10-19 | 腾讯科技（深圳）有限公司 | Image object recognition methods and device, storage medium |
| CN108921140A (en) * | 2018-08-07 | 2018-11-30 | 安徽云森物联网科技有限公司 | Pedestrian's recognition methods again |
| CN109034109A (en) * | 2018-08-16 | 2018-12-18 | 新智数字科技有限公司 | A kind of pedestrian based on clustering algorithm recognition methods and device again |
| CN109086690A (en) * | 2018-07-13 | 2018-12-25 | 北京旷视科技有限公司 | Image characteristic extracting method, target identification method and corresponding intrument |
| CN109101913A (en) * | 2018-08-01 | 2018-12-28 | 北京飞搜科技有限公司 | Pedestrian recognition methods and device again |
| CN109241902A (en) * | 2018-08-30 | 2019-01-18 | 北京航空航天大学 | A kind of landslide detection method based on multi-scale feature fusion |
| CN109271895A (en) * | 2018-08-31 | 2019-01-25 | 西安电子科技大学 | Pedestrian's recognition methods again based on Analysis On Multi-scale Features study and Image Segmentation Methods Based on Features |
| CN109345506A (en) * | 2018-08-23 | 2019-02-15 | 中国科学院合肥物质科学研究院 | A kind of hot spot based on convolutional neural networks and MARFE automatic testing method |
| CN109492610A (en) * | 2018-11-27 | 2019-03-19 | 广东工业大学 | A kind of pedestrian recognition methods, device and readable storage medium storing program for executing again |
| CN109740413A (en) * | 2018-11-14 | 2019-05-10 | 平安科技（深圳）有限公司 | Pedestrian recognition methods, device, computer equipment and computer storage medium again |
| CN110082283A (en) * | 2019-05-23 | 2019-08-02 | 山东科技大学 | A kind of Atmospheric particulates SEM image recognition methods and system |
| CN110148052A (en) * | 2019-04-17 | 2019-08-20 | 深圳壹账通智能科技有限公司 | Management-control method, device, computer equipment and storage medium after businessman borrows |
| CN110175506A (en) * | 2019-04-08 | 2019-08-27 | 复旦大学 | Pedestrian based on parallel dimensionality reduction convolutional neural networks recognition methods and device again |
| CN110175904A (en) * | 2019-04-15 | 2019-08-27 | 深圳壹账通智能科技有限公司 | Detection method, system, equipment and the storage medium of shop volume of the flow of passengers level ground effect |
| CN110516787A (en) * | 2019-07-15 | 2019-11-29 | 杭州电子科技大学 | Deep learning network regularization constraint method based on easy dtex sign drop policy |
| CN110543817A (en) * | 2019-07-25 | 2019-12-06 | 北京大学 | Pedestrian re-identification method based on posture guidance feature learning |
| CN110705431A (en) * | 2019-09-26 | 2020-01-17 | 中国人民解放军陆军炮兵防空兵学院 | Video saliency region detection method and system based on depth C3D feature |
| CN110827208A (en) * | 2019-09-19 | 2020-02-21 | 重庆特斯联智慧科技股份有限公司 | General pooling enhancement method, device, equipment and medium for convolutional neural network |
| CN110909701A (en) * | 2019-11-28 | 2020-03-24 | 北京百度网讯科技有限公司 | Pedestrian feature extraction method, device, equipment and medium |
| CN111044045A (en) * | 2019-12-09 | 2020-04-21 | 中国科学院深圳先进技术研究院 | Navigation method and device based on neural network and terminal equipment |
| CN111144294A (en) * | 2019-12-26 | 2020-05-12 | 上海眼控科技股份有限公司 | Target identification method and device, computer equipment and readable storage medium |
| CN111191561A (en) * | 2019-12-25 | 2020-05-22 | 北京迈格威科技有限公司 | Method, apparatus and computer storage medium for re-identification of non-motor vehicles |
| CN111291606A (en) * | 2019-04-16 | 2020-06-16 | 北京潼荔科技有限公司 | Scene self-adaptive target recognition artificial intelligence method and system based on edge calculation |
| CN111428612A (en) * | 2020-03-19 | 2020-07-17 | 深圳力维智联技术有限公司 | Pedestrian re-identification method, terminal, device and storage medium |
| CN111523470A (en) * | 2020-04-23 | 2020-08-11 | 苏州浪潮智能科技有限公司 | Feature fusion block, convolutional neural network, pedestrian re-identification method and related equipment |
| CN111814857A (en) * | 2020-06-29 | 2020-10-23 | 浙江大华技术股份有限公司 | Target re-identification method, network training method thereof and related device |
| CN111898732A (en) * | 2020-06-30 | 2020-11-06 | 江苏省特种设备安全监督检验研究院 | Ultrasonic ranging compensation method based on deep convolutional neural network |
| CN111914668A (en) * | 2020-07-08 | 2020-11-10 | 浙江大华技术股份有限公司 | Pedestrian re-identification method, device and system based on image enhancement technology |
| CN112347957A (en) * | 2020-11-12 | 2021-02-09 | 广联达科技股份有限公司 | Pedestrian re-identification method and device, computer equipment and storage medium |
| CN112784648A (en) * | 2019-11-07 | 2021-05-11 | 中国科学技术大学 | Method and device for optimizing feature extraction of pedestrian re-identification system of video |
| CN113486815A (en) * | 2021-07-09 | 2021-10-08 | 山东力聚机器人科技股份有限公司 | Pedestrian re-identification system and method, computer equipment and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130343642A1 (en) * | 2012-06-21 | 2013-12-26 | Siemens Corporation | Machine-learnt person re-identification |
| CN106203533A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | The degree of depth based on combined training study face verification method |
| CN106709449A (en) * | 2016-12-22 | 2017-05-24 | 深圳市深网视界科技有限公司 | Pedestrian re-recognition method and system based on deep learning and reinforcement learning |
| CN106778464A (en) * | 2016-11-09 | 2017-05-31 | 深圳市深网视界科技有限公司 | A kind of pedestrian based on deep learning recognition methods and device again |
| CN106780906A (en) * | 2016-12-28 | 2017-05-31 | 北京品恩科技股份有限公司 | A kind of testimony of a witness unification recognition methods and system based on depth convolutional neural networks |
| CN106845330A (en) * | 2016-11-17 | 2017-06-13 | 北京品恩科技股份有限公司 | A kind of training method of the two-dimension human face identification model based on depth convolutional neural networks |
| US20170169315A1 (en) * | 2015-12-15 | 2017-06-15 | Sighthound, Inc. | Deeply learned convolutional neural networks (cnns) for object localization and classification |
| CN107145845A (en) * | 2017-04-26 | 2017-09-08 | 中山大学 | The pedestrian detection method merged based on deep learning and multi-characteristic points |
-
2017
- 2017-10-26 CN CN201711017356.6A patent/CN107657249A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130343642A1 (en) * | 2012-06-21 | 2013-12-26 | Siemens Corporation | Machine-learnt person re-identification |
| US20170169315A1 (en) * | 2015-12-15 | 2017-06-15 | Sighthound, Inc. | Deeply learned convolutional neural networks (cnns) for object localization and classification |
| CN106203533A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | The degree of depth based on combined training study face verification method |
| CN106778464A (en) * | 2016-11-09 | 2017-05-31 | 深圳市深网视界科技有限公司 | A kind of pedestrian based on deep learning recognition methods and device again |
| CN106845330A (en) * | 2016-11-17 | 2017-06-13 | 北京品恩科技股份有限公司 | A kind of training method of the two-dimension human face identification model based on depth convolutional neural networks |
| CN106709449A (en) * | 2016-12-22 | 2017-05-24 | 深圳市深网视界科技有限公司 | Pedestrian re-recognition method and system based on deep learning and reinforcement learning |
| CN106780906A (en) * | 2016-12-28 | 2017-05-31 | 北京品恩科技股份有限公司 | A kind of testimony of a witness unification recognition methods and system based on depth convolutional neural networks |
| CN107145845A (en) * | 2017-04-26 | 2017-09-08 | 中山大学 | The pedestrian detection method merged based on deep learning and multi-characteristic points |
Non-Patent Citations (2)
| Title |
|---|
| JIAWEI LIU ET AL: "Multi-scale triplet CNN for person re-identification", 《PROCEEDINGS OF THE 24TH ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA》 * |
| XUELIN QIAN ET AL: "Multi-scale Deep Learning Architectures for Person Re-identification", 《ARXIV》 * |
Cited By (44)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108491880A (en) * | 2018-03-23 | 2018-09-04 | 西安电子科技大学 | Object classification based on neural network and position and orientation estimation method |
| CN108681743B (en) * | 2018-04-16 | 2019-12-06 | 腾讯科技（深圳）有限公司 | Image object recognition method and device and storage medium |
| CN108681743A (en) * | 2018-04-16 | 2018-10-19 | 腾讯科技（深圳）有限公司 | Image object recognition methods and device, storage medium |
| US11227182B2 (en) | 2018-04-16 | 2022-01-18 | Tencent Technology (Shenzhen) Company Limited | Method, apparatus, and storage medium for recognizing image object |
| CN109086690A (en) * | 2018-07-13 | 2018-12-25 | 北京旷视科技有限公司 | Image characteristic extracting method, target identification method and corresponding intrument |
| CN109101913A (en) * | 2018-08-01 | 2018-12-28 | 北京飞搜科技有限公司 | Pedestrian recognition methods and device again |
| CN108921140A (en) * | 2018-08-07 | 2018-11-30 | 安徽云森物联网科技有限公司 | Pedestrian's recognition methods again |
| CN109034109B (en) * | 2018-08-16 | 2021-03-23 | 新智数字科技有限公司 | Pedestrian re-identification method and device based on clustering algorithm |
| CN109034109A (en) * | 2018-08-16 | 2018-12-18 | 新智数字科技有限公司 | A kind of pedestrian based on clustering algorithm recognition methods and device again |
| CN109345506A (en) * | 2018-08-23 | 2019-02-15 | 中国科学院合肥物质科学研究院 | A kind of hot spot based on convolutional neural networks and MARFE automatic testing method |
| CN109241902A (en) * | 2018-08-30 | 2019-01-18 | 北京航空航天大学 | A kind of landslide detection method based on multi-scale feature fusion |
| CN109271895A (en) * | 2018-08-31 | 2019-01-25 | 西安电子科技大学 | Pedestrian's recognition methods again based on Analysis On Multi-scale Features study and Image Segmentation Methods Based on Features |
| CN109740413A (en) * | 2018-11-14 | 2019-05-10 | 平安科技（深圳）有限公司 | Pedestrian recognition methods, device, computer equipment and computer storage medium again |
| CN109492610A (en) * | 2018-11-27 | 2019-03-19 | 广东工业大学 | A kind of pedestrian recognition methods, device and readable storage medium storing program for executing again |
| CN110175506A (en) * | 2019-04-08 | 2019-08-27 | 复旦大学 | Pedestrian based on parallel dimensionality reduction convolutional neural networks recognition methods and device again |
| CN110175904A (en) * | 2019-04-15 | 2019-08-27 | 深圳壹账通智能科技有限公司 | Detection method, system, equipment and the storage medium of shop volume of the flow of passengers level ground effect |
| CN111291606A (en) * | 2019-04-16 | 2020-06-16 | 北京潼荔科技有限公司 | Scene self-adaptive target recognition artificial intelligence method and system based on edge calculation |
| CN110148052A (en) * | 2019-04-17 | 2019-08-20 | 深圳壹账通智能科技有限公司 | Management-control method, device, computer equipment and storage medium after businessman borrows |
| CN110082283B (en) * | 2019-05-23 | 2021-12-14 | 山东科技大学 | Atmospheric particulate SEM image recognition method and system |
| CN110082283A (en) * | 2019-05-23 | 2019-08-02 | 山东科技大学 | A kind of Atmospheric particulates SEM image recognition methods and system |
| CN110516787A (en) * | 2019-07-15 | 2019-11-29 | 杭州电子科技大学 | Deep learning network regularization constraint method based on easy dtex sign drop policy |
| CN110516787B (en) * | 2019-07-15 | 2021-04-09 | 杭州电子科技大学 | Pedestrian re-identification method based on network regularization constraint of easily-separable feature discarding |
| CN110543817A (en) * | 2019-07-25 | 2019-12-06 | 北京大学 | Pedestrian re-identification method based on posture guidance feature learning |
| CN110827208A (en) * | 2019-09-19 | 2020-02-21 | 重庆特斯联智慧科技股份有限公司 | General pooling enhancement method, device, equipment and medium for convolutional neural network |
| CN110705431A (en) * | 2019-09-26 | 2020-01-17 | 中国人民解放军陆军炮兵防空兵学院 | Video saliency region detection method and system based on depth C3D feature |
| CN110705431B (en) * | 2019-09-26 | 2022-03-15 | 中国人民解放军陆军炮兵防空兵学院 | Video saliency region detection method and system based on depth C3D feature |
| CN112784648A (en) * | 2019-11-07 | 2021-05-11 | 中国科学技术大学 | Method and device for optimizing feature extraction of pedestrian re-identification system of video |
| CN112784648B (en) * | 2019-11-07 | 2022-09-06 | 中国科学技术大学 | Method and device for optimizing feature extraction of pedestrian re-identification system of video |
| CN110909701B (en) * | 2019-11-28 | 2023-03-24 | 北京百度网讯科技有限公司 | Pedestrian feature extraction method, device, equipment and medium |
| CN110909701A (en) * | 2019-11-28 | 2020-03-24 | 北京百度网讯科技有限公司 | Pedestrian feature extraction method, device, equipment and medium |
| CN111044045A (en) * | 2019-12-09 | 2020-04-21 | 中国科学院深圳先进技术研究院 | Navigation method and device based on neural network and terminal equipment |
| CN111191561A (en) * | 2019-12-25 | 2020-05-22 | 北京迈格威科技有限公司 | Method, apparatus and computer storage medium for re-identification of non-motor vehicles |
| CN111144294A (en) * | 2019-12-26 | 2020-05-12 | 上海眼控科技股份有限公司 | Target identification method and device, computer equipment and readable storage medium |
| CN111428612B (en) * | 2020-03-19 | 2023-08-15 | 深圳力维智联技术有限公司 | Pedestrian re-identification method, terminal, device and storage medium |
| CN111428612A (en) * | 2020-03-19 | 2020-07-17 | 深圳力维智联技术有限公司 | Pedestrian re-identification method, terminal, device and storage medium |
| CN111523470A (en) * | 2020-04-23 | 2020-08-11 | 苏州浪潮智能科技有限公司 | Feature fusion block, convolutional neural network, pedestrian re-identification method and related equipment |
| CN111523470B (en) * | 2020-04-23 | 2022-11-18 | 苏州浪潮智能科技有限公司 | Pedestrian re-identification method, device, equipment and medium |
| CN111814857A (en) * | 2020-06-29 | 2020-10-23 | 浙江大华技术股份有限公司 | Target re-identification method, network training method thereof and related device |
| CN111898732B (en) * | 2020-06-30 | 2023-06-20 | 江苏省特种设备安全监督检验研究院 | Ultrasonic ranging compensation method based on deep convolutional neural network |
| CN111898732A (en) * | 2020-06-30 | 2020-11-06 | 江苏省特种设备安全监督检验研究院 | Ultrasonic ranging compensation method based on deep convolutional neural network |
| CN111914668A (en) * | 2020-07-08 | 2020-11-10 | 浙江大华技术股份有限公司 | Pedestrian re-identification method, device and system based on image enhancement technology |
| CN112347957A (en) * | 2020-11-12 | 2021-02-09 | 广联达科技股份有限公司 | Pedestrian re-identification method and device, computer equipment and storage medium |
| CN113486815A (en) * | 2021-07-09 | 2021-10-08 | 山东力聚机器人科技股份有限公司 | Pedestrian re-identification system and method, computer equipment and storage medium |
| CN113486815B (en) * | 2021-07-09 | 2022-10-21 | 山东力聚机器人科技股份有限公司 | Pedestrian re-identification system and method, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107657249A (en) | Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again | |
| CN111898547B (en) | Training method, device, equipment and storage medium of face recognition model | |
| CN108509859B (en) | Non-overlapping area pedestrian tracking method based on deep neural network | |
| CN110728209B (en) | Gesture recognition method and device, electronic equipment and storage medium | |
| CN106485215B (en) | Face shielding detection method based on deep convolutional neural network | |
| CN106960195B (en) | Crowd counting method and device based on deep learning | |
| CN111274916B (en) | Face recognition method and face recognition device | |
| CN106845487B (en) | End-to-end license plate identification method | |
| US8401292B2 (en) | Identifying high saliency regions in digital images | |
| CN107844753A (en) | Pedestrian in video image recognition methods, device, storage medium and processor again | |
| CN109284733B (en) | Shopping guide negative behavior monitoring method based on yolo and multitask convolutional neural network | |
| CN111611874B (en) | Face mask wearing detection method based on ResNet and Canny | |
| CN108090433A (en) | Face identification method and device, storage medium, processor | |
| CN105469376B (en) | The method and apparatus for determining picture similarity | |
| CN112861690B (en) | Multi-method fused remote sensing image change detection method and system | |
| CN111160249A (en) | Multi-class target detection method of optical remote sensing image based on cross-scale feature fusion | |
| CN107220635A (en) | Human face in-vivo detection method based on many fraud modes | |
| CN104992223A (en) | Intensive population estimation method based on deep learning | |
| CN112016464A (en) | Method and device for detecting face shielding, electronic equipment and storage medium | |
| CN105975929A (en) | Fast pedestrian detection method based on aggregated channel features | |
| CN107609512A (en) | A kind of video human face method for catching based on neutral net | |
| CN107545249A (en) | A kind of population ages' recognition methods and device | |
| CN111582092B (en) | Pedestrian abnormal behavior detection method based on human skeleton | |
| CN110287889A (en) | A kind of method and device of identification | |
| CN110263920A (en) | Convolutional neural networks model and its training method and device, method for inspecting and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180202 |
|
| WD01 | Invention patent application deemed withdrawn after publication |