CN111582447B - Closed loop detection method based on multiple network characteristics - Google Patents
Closed loop detection method based on multiple network characteristics Download PDFInfo
- Publication number
- CN111582447B CN111582447B CN202010365392.7A CN202010365392A CN111582447B CN 111582447 B CN111582447 B CN 111582447B CN 202010365392 A CN202010365392 A CN 202010365392A CN 111582447 B CN111582447 B CN 111582447B
- Authority
- CN
- China
- Prior art keywords
- image
- neural network
- loop detection
- closed
- descriptor
- 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.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 43
- 238000003062 neural network model Methods 0.000 claims abstract description 48
- 238000000034 method Methods 0.000 claims abstract description 22
- 230000009467 reduction Effects 0.000 claims abstract description 17
- 239000011159 matrix material Substances 0.000 claims description 16
- 238000012549 training Methods 0.000 claims description 7
- 230000003044 adaptive effect Effects 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000013526 transfer learning Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 abstract description 3
- 238000013527 convolutional neural network Methods 0.000 abstract description 3
- 230000007613 environmental effect Effects 0.000 abstract description 2
- 239000010410 layer Substances 0.000 description 29
- 230000000007 visual effect Effects 0.000 description 8
- 238000013135 deep learning Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000013507 mapping Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000004807 localization Effects 0.000 description 3
- 239000000284 extract Substances 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 244000304337 Cuminum cyminum Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 201000011243 gastrointestinal stromal tumor Diseases 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
Images
Classifications
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a closed-loop detection method based on multiple network characteristics, which has better robust performance under the condition of environmental change, adopts the multiple network characteristics as an image characteristic descriptor, and represents an image by utilizing the characteristic combination mode of different network output layers of different neural network models, so that the image descriptor can carry richer and more accurate information, thereby improving the performance of the whole closed-loop detection algorithm. Compared with other methods for closed-loop detection by using a convolutional neural network, the method combines the characteristics of different levels of a plurality of neural network models, integrates the low-level and high-level image information of the different neural network models, enriches the image representation, and ensures the real-time performance of the method by performing feature dimension reduction operation before feature cascade.
Description
Technical Field
The invention belongs to the technical field of visual SLAM, and particularly relates to a design of a closed-loop detection method based on various network characteristics.
Background
SLAM is an abbreviation for "Simultaneous Localization And Mapping" And can be translated into Simultaneous Localization And Mapping. The probabilistic SLAM problem (the probabilistic SLAM project) originated at the IEEE Robotics and Automation Conference in 1986, and researchers hoped to apply the estimation-theoretical methods to the mapping and localization problems. SLAM was first applied in the field of robots, with the goal of building a map of the surrounding environment from sensor data in real time without any prior knowledge, while inferring its own position from this map. In recent years, visual SLAM has found wide application in the fields of robotics, autonomous cars/drones, and AR/VR.
A complete visual SLAM system requires several key modules such as data processing, visual odometry, back-end optimization, mapping and closed-loop detection. Closed loop detection is indeed a crucial part for mobile robots based on various sensors. A good visual SLAM system requires closed loop detection techniques to help it reduce the cumulative error associated with visual odometers. When the closed loop is detected, the obtained information is provided for the back end to be processed, the accumulated error in the moving process of the robot can be obviously reduced, and therefore the accuracy of the back end optimization processing is improved, and therefore the closed loop detection has very important significance for improving the accuracy and the robustness of the whole SLAM system.
For closed-loop detection, feature extraction is a very critical part, and is mainly implemented in two ways, namely, a traditional way based on various traditional image descriptor technologies. For example, local image features (SIFT, bag-of-Words, fisher vector, and VLAD for SURF and ORB) are used. Other methods use global image features, such as GIST. The famous FAB-MAP proposed by Cummins et al extracts SURF characteristics and constructs a visual dictionary (BOW model), and the robot searches the acquired characters in the acquired images by using the visual dictionary in the running process to obtain the similarity between two places and calculate the probability of closed loop to realize closed loop detection. The other is a neural network-based mode, and because the features extracted in the traditional mode are designed according to the professional knowledge and the prior knowledge of people, the images cannot be represented well, and the features have certain limitations and poor generalization performance. In recent years, deep learning has been rapidly rising, and great progress has been made in recent years. Recent studies have demonstrated significant success of deep learning in various computer vision tasks, further prompting researchers to begin developing studies in closed-loop detection based on neural networks. Currently, researchers have used two deep learning methods in closed-loop detection. One is the manner of an auto-encoder, which is trained with the auto-encoder to learn the feature representation and find the closed loop by using a similarity or difference matrix. Another is that in environments where lighting changes, pre-trained neural network models can extract features and outperform the latest manually extracted features. For example, gao X et al propose training stacked autoencoders to learn feature representations. On the basis of a sparse automatic encoder, hu Han et al change an activation function of an output layer into an identity function, and the problem that input samples need to be scaled and are not suitable for color images is solved. In the same year, yi Hou et al propose to use a convolutional neural network for the first time to realize closed-loop detection, and have a better detection effect. In 2018, xia Y et al compare the deep learning classical frame with the traditional closed-loop detection method, and obtain the result that the deep learning method has better closed-loop detection performance than the traditional algorithm in terms of both precision and processing time. However, the existing closed-loop detection method only uses a single-layer network representation image of a single neural network, so that the image representation is not rich, and the algorithm performance is further influenced.
Disclosure of Invention
The invention aims to provide a closed-loop detection method based on various network characteristics, aiming at the problems that the image representation is not abundant and the algorithm performance is influenced because the image representation is only represented by a single network of a single neural network in the existing closed-loop detection method.
The technical scheme of the invention is as follows: the closed loop detection method based on various network characteristics comprises the following steps:
s1, obtaining a plurality of different neural network models of the adaptive closed-loop detection system, and combining different network characteristics in the neural network models to serve as image descriptors.
And S2, carrying out data preprocessing on the image descriptor.
And S3, calculating a similarity matrix of the image data set according to the data preprocessing result.
And S4, carrying out closed-loop detection according to the similarity matrix.
Further, step S1 comprises the following sub-steps:
s11, training different neural network models, and determining the optimal input, output, network layer number and activation function of each neural network model, or training partial parameters of the existing neural network model by using a transfer learning method to obtain a plurality of different neural network models of the adaptive closed-loop detection system.
And S12, finding out the optimal combination mode of different output layers in each neural network model in sequence, and recording the output layer position of the optimal combination mode in each neural network model.
And S13, listing the positions of the output layers in the optimal combination mode in each neural network model, and randomly combining the optimal output layers under different neural network models to be used for representing images to be used as image descriptors.
Further, step S2 comprises the following sub-steps:
and S21, performing dimension reduction processing on the image descriptor by adopting a dimension reduction algorithm.
And S22, combining the image descriptors after the dimension reduction.
And S23, normalizing the combined image descriptor.
Further, the specific method for combining the image descriptors after the dimension reduction in step S22 is as follows: and distributing the proportion of different neural network models in the image descriptors in a direct cascade mode to realize the combination of the image descriptors.
Further, the specific method for combining the image descriptors after the dimension reduction in step S22 is as follows: and adjusting the proportion of the network layers of different neural network models, finding out proper parameters and obtaining the combined image descriptor.
Further, the formula of the normalization process in step S23 is:
where img proc represents the normalized image descriptor, img represents the image feature matrix after feature concatenation,
the L2 norm representing img and d the output dimension of a certain layer of the neural network model.
Further, the calculation formula of the similarity matrix in step S3 is:
where Sim (i, j) denotes the ith image l i And j image l j The similarity matrix of (a) is obtained,
representing the ith image l i Normalized image descriptor->
Representing the j-th image l j The normalized image descriptor is then used to determine the image descriptor,
representing the ith image l i And j image l j The difference size of (a), sim (i, j) is E [0,1]。
Further, step S4 specifically includes: and setting a threshold Th, judging to be a closed loop if Sim (i, j) is more than or equal to Th, otherwise, judging to be a non-closed loop, and realizing closed loop detection.
The invention has the beneficial effects that: the invention has better robust performance under the condition of environmental change, and adopts a plurality of network characteristics as the image characteristic descriptor, on one hand, because the emphasis points of the image characteristics output by different neural network models are different, but one image can be well represented, on the other hand, the characteristics of the image under the model can be well represented by considering only one layer or a plurality of layers of the same network model, therefore, the invention utilizes the characteristic combination mode of different network output layers of different neural network models to represent the image, so that the image descriptor can carry richer and more accurate information, thereby improving the performance of the whole closed-loop detection algorithm. Compared with other methods for closed-loop detection by using a convolutional neural network, the method combines the characteristics of different levels of a plurality of neural network models, integrates the low-level and high-level image information of the different neural network models, enriches the image representation, and ensures the real-time performance of the method by performing feature dimension reduction operation before feature cascade.
Drawings
Fig. 1 is a general block diagram of a closed-loop detection method based on various network characteristics according to an embodiment of the present invention.
Fig. 2 is a flowchart of a closed-loop detection method based on multiple network features according to an embodiment of the present invention.
Fig. 3 is a schematic diagram of an AlexNet network structure according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of a VGG16 network structure according to an embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It is to be understood that the embodiments shown and described in the drawings are merely exemplary and are intended to illustrate the principles and spirit of the invention, rather than to limit the scope of the invention.
The embodiment of the invention provides a closed loop detection method based on various network characteristics, which comprises the following steps S1-S4 as shown in fig. 1 and fig. 2 together:
s1, obtaining a plurality of different neural network models of the adaptive closed-loop detection system, and combining different network characteristics in the neural network models to serve as an image descriptor, wherein the descriptor is a data structure for characterizing the characteristics.
Step S1 includes the following substeps S11-S13:
s11, training different neural network models, and determining the optimal input, output, network layer number and activation function of each neural network model, or performing partial parameter training on the existing neural network model by using a transfer learning method to obtain a plurality of different neural network models of the adaptive closed-loop detection system.
In the embodiment of the invention, the existing neural network model adopts a pre-trained VGG16 network model and an AlexNet network model, wherein the structure of the AlexNet network is shown in FIG. 3, and the structure of the VGG16 network is shown in FIG. 4.
S12, finding out the optimal combination mode of different output layers in each neural network model in sequence, and recording the output layer position of the optimal combination mode in each neural network model, wherein the output layer position represents typical and appropriate image characteristics extracted from each neural network model.
In the embodiment of the invention, the optimal combination of AlexNet network structures is pool2 and pool5, and the optimal combination of VGG16 network structures is block2_ conv2 and block5_ conv3.
And S13, listing the position of an output layer in the optimal combination mode in each neural network model, and randomly combining the optimal output layers under different neural network models to represent images to serve as image descriptors, so that the image descriptors are more abundant in expression and better in robustness.
In the embodiment of the present invention, for example, a pool2 layer of AlexNet and a block5_ conv3 of VGG16 are combined to form an image feature 1, a pool5 layer of AlexNet and a block2_ conv2 layer of VGG16 are combined to form an image feature 2, an optimal combination mode is selected from different combinations to be used as a descriptor of an input image, and it is assumed that the image feature 1 is an optimal image feature representation, that is, the image feature 1 is selected as an image descriptor.
And S2, carrying out data preprocessing on the image descriptor.
Step S2 includes the following substeps S21-S23:
and S21, performing dimensionality reduction processing on the image descriptor by adopting a dimensionality reduction algorithm, so as to realize better real-time performance, reduce characteristic redundancy and improve algorithm performance.
In the embodiment of the present invention, assuming that the sample size is m,
the ith image feature extracted from the pool5 layer of the AlexNet model is represented, d1 represents the dimension of the feature extracted from the pool5 layer of the AlexNet model, and lay1 represents the name of the output layer corresponding to the current network model, here, pool5. Then the current sample
And representing a new image descriptor of the ith image after the dimension reduction algorithm. The descriptors of the same image obtained from different levels are uniformly reduced to 100 dimensions, then ^ er>
In the same way, assume->
Representing the feature extracted from the block5_ conv3 layer of the VGG16 model to the ith image, d2 representing the dimension of the feature extracted from the block5_ conv3 layer of the VGG16 model to the image, lay2 representing the name of the output layer corresponding to the current network model, here, block5_ conv3, then the feature after dimension reduction is represented as ^ 5_ conv3>
And S22, combining the image descriptors after the dimension reduction.
In the embodiment of the invention, two specific schemes are combined for the image descriptors after dimension reduction:
(1) And distributing the proportion of different neural network models in the image descriptors in a direct cascade mode to realize the combination of the image descriptors.
(2) And adjusting the proportion of the network layers of different neural network models, finding out proper parameters and obtaining the combined image descriptor.
Taking the first scheme as an example, the image descriptors are directly composed of different network output layer characteristics according to the same proportion, namely the image descriptors img ∈ R at this time 200 。
S23, normalizing the combined image descriptor to reduce the data range to a range of [0,1], and normalizing the data, wherein the normalization formula is as follows:
where img proc represents the normalized image descriptor, img represents the image feature matrix after feature concatenation,
and the L2 norm of img is represented, the normalization of the image descriptor can be realized by adopting other norm forms, and d represents the output dimension of a certain layer of the neural network model.
And S3, calculating a similarity matrix of the image data set according to the data preprocessing result.
Common similarity algorithms include an Euclidean distance method, a cosine distance method and the like, a similarity matrix is optimized according to actual conditions, and algorithm performance is improved.
Where Sim (i, j) denotes the ith image l i And j image l j The similarity matrix of (a) is obtained,
representing the ith image l i The normalized image descriptor->
Representing the j-th image l j The normalized image descriptor is then used to determine the image descriptor,
representing the ith image l i And j image l j The difference size of (a), sim (i, j) is E [0,1]。
And S4, carrying out closed-loop detection according to the similarity matrix.
And setting a threshold Th, judging to be a closed loop if Sim (i, j) is more than or equal to Th, otherwise, judging to be a non-closed loop, and realizing closed loop detection.
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.
Claims (4)
1. The closed loop detection method based on various network characteristics is characterized by comprising the following steps:
s1, obtaining a plurality of different neural network models of the adaptive closed-loop detection system, and combining different network characteristics in the neural network models to serve as image descriptors;
s2, carrying out data preprocessing on the image descriptor;
s3, calculating a similarity matrix of the image data set according to the data preprocessing result;
s4, performing closed-loop detection according to the similarity matrix;
the step S1 comprises the following sub-steps:
s11, training different neural network models, and determining the optimal input, output, network layer number and activation function of each neural network model, or training partial parameters of the existing neural network model by using a transfer learning method to obtain a plurality of different neural network models of the adaptive closed-loop detection system;
s12, finding out the optimal combination mode of different output layers in each neural network model in sequence, and recording the output layer position of the optimal combination mode in each neural network model;
s13, listing the position of an output layer in an optimal combination mode in each neural network model, and randomly combining the optimal output layers under different neural network models to be used for representing images to be used as image descriptors;
the step S2 comprises the following sub-steps:
s21, performing dimensionality reduction processing on the image descriptor by adopting a dimensionality reduction algorithm;
s22, combining the image descriptors after dimension reduction;
s23, normalizing the combined image descriptor;
the specific method for combining the image descriptors after the dimension reduction in the step S22 is as follows: and adjusting the proportion of the network layers of different neural network models, finding out proper parameters and obtaining the combined image descriptor.
2. The closed-loop detection method according to claim 1, wherein the formula of the normalization process in step S23 is:
where img proc represents the normalized image descriptor, img represents the image feature matrix after feature concatenation,
the L2 norm representing img and d represents the output dimension of a certain layer of the neural network model.
3. The closed-loop detection method according to claim 1, wherein the similarity matrix in step S3 is calculated by the following formula:
where Sim (i, j) denotes the ith image l i And j image l j The similarity matrix of (a) is obtained,
representing the ith image l i The normalized image descriptor->
Representing the j-th image l j The normalized image descriptor is then used to determine the image descriptor,
representing the ith image l i And j image l j The difference size of (a), sim (i, j) is E [0,1]。
4. The closed-loop detection method according to claim 3, wherein the step S4 specifically comprises: and setting a threshold Th, judging to be a closed loop if Sim (i, j) is more than or equal to Th, otherwise, judging to be a non-closed loop, and realizing closed loop detection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010365392.7A CN111582447B (en) | 2020-04-30 | 2020-04-30 | Closed loop detection method based on multiple network characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010365392.7A CN111582447B (en) | 2020-04-30 | 2020-04-30 | Closed loop detection method based on multiple network characteristics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111582447A CN111582447A (en) | 2020-08-25 |
| CN111582447B true CN111582447B (en) | 2023-04-07 |
Family
ID=72111993
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010365392.7A Active CN111582447B (en) | 2020-04-30 | 2020-04-30 | Closed loop detection method based on multiple network characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111582447B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110166154A (en) * | 2019-06-13 | 2019-08-23 | 桂林电子科技大学 | A kind of software radio spectrum monitoring knowledge method for distinguishing neural network based |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2554633B (en) * | 2016-06-24 | 2020-01-22 | Imperial College Sci Tech & Medicine | Detecting objects in video data |
| CN107833220B (en) * | 2017-11-28 | 2021-06-11 | 河海大学常州校区 | Fabric defect detection method based on deep convolutional neural network and visual saliency |
| CN109443382B (en) * | 2018-10-22 | 2022-05-17 | 北京工业大学 | Visual SLAM closed loop detection method based on feature extraction and dimension reduction neural network |
| CN109341694A (en) * | 2018-11-12 | 2019-02-15 | 哈尔滨理工大学 | A kind of autonomous positioning air navigation aid of mobile sniffing robot |
| CN110188775B (en) * | 2019-05-28 | 2020-06-26 | 创意信息技术股份有限公司 | Image content description automatic generation method based on joint neural network model |
| CN110880010A (en) * | 2019-07-05 | 2020-03-13 | 电子科技大学 | Visual SLAM closed loop detection algorithm based on convolutional neural network |
| CN110555881A (en) * | 2019-08-29 | 2019-12-10 | 桂林电子科技大学 | Visual SLAM testing method based on convolutional neural network |
| CN110533661A (en) * | 2019-09-04 | 2019-12-03 | 电子科技大学 | Adaptive real-time closed-loop detection method based on characteristics of image cascade |
| CN110781790A (en) * | 2019-10-19 | 2020-02-11 | 北京工业大学 | Visual SLAM closed loop detection method based on convolutional neural network and VLAD |
| CN110852327A (en) * | 2019-11-07 | 2020-02-28 | 首都师范大学 | Image processing method, image processing device, electronic equipment and storage medium |
-
2020
- 2020-04-30 CN CN202010365392.7A patent/CN111582447B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110166154A (en) * | 2019-06-13 | 2019-08-23 | 桂林电子科技大学 | A kind of software radio spectrum monitoring knowledge method for distinguishing neural network based |
Non-Patent Citations (1)
| Title |
|---|
| 杨孟军 ; 苏成悦 ; 陈静 ; 张洁鑫 ; .基于卷积神经网络的视觉闭环检测研究.广东工业大学学报.2018,(第05期),第35-41页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111582447A (en) | 2020-08-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110414462B (en) | Unsupervised cross-domain pedestrian re-identification method and system | |
| CN110363816A (en) | A kind of mobile robot environment semanteme based on deep learning builds drawing method | |
| CN110781790A (en) | Visual SLAM closed loop detection method based on convolutional neural network and VLAD | |
| CN111079847A (en) | Remote sensing image automatic labeling method based on deep learning | |
| CN111274958B (en) | Pedestrian re-identification method and system with network parameter self-correction function | |
| CN115375951B (en) | Small sample hyperspectral image classification method based on primitive migration network | |
| CN111126563B (en) | Target identification method and system based on space-time data of twin network | |
| CN104517289A (en) | Indoor scene positioning method based on hybrid camera | |
| CN110533661A (en) | Adaptive real-time closed-loop detection method based on characteristics of image cascade | |
| Bai et al. | Matching-range-constrained real-time loop closure detection with CNNs features | |
| CN113065409A (en) | Unsupervised pedestrian re-identification method based on camera distribution difference alignment constraint | |
| Wang et al. | Online visual place recognition via saliency re-identification | |
| CN112084895A (en) | Pedestrian re-identification method based on deep learning | |
| CN107452374A (en) | Based on unidirectional from the various visual angles language identification method for marking auxiliary information | |
| Tsintotas et al. | The revisiting problem in simultaneous localization and mapping | |
| Yan et al. | An incremental intelligent object recognition system based on deep learning | |
| CN111582447B (en) | Closed loop detection method based on multiple network characteristics | |
| CN114444571A (en) | Sonar target individual identification method for autonomous learning | |
| Gu et al. | An improved loop closure detection for RatSLAM | |
| CN114973099A (en) | Intelligent object searching method and system based on traceable target identification | |
| CN112487927B (en) | Method and system for realizing indoor scene recognition based on object associated attention | |
| CN115187614A (en) | Real-time simultaneous positioning and mapping method based on STDC semantic segmentation network | |
| CN114358068B (en) | Space potential measurement method and system for neural network small sample learning | |
| CN113033669B (en) | Visual scene recognition method based on learnable feature map filtering and graph annotation meaning network | |
| Zhang et al. | Motion Field Consensus with Locality Preservation: A Geometric Confirmation Strategy for Loop Closure Detection |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |