CN114550036A - Rapid search method and system for optimal cascade configuration of video target detection - Google Patents
Rapid search method and system for optimal cascade configuration of video target detection Download PDFInfo
- Publication number
- CN114550036A CN114550036A CN202210147889.0A CN202210147889A CN114550036A CN 114550036 A CN114550036 A CN 114550036A CN 202210147889 A CN202210147889 A CN 202210147889A CN 114550036 A CN114550036 A CN 114550036A
- Authority
- CN
- China
- Prior art keywords
- data set
- video data
- video
- cascade
- scheme
- 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.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 61
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 52
- 238000001914 filtration Methods 0.000 claims abstract description 41
- 238000004364 calculation method Methods 0.000 claims abstract description 27
- 238000010845 search algorithm Methods 0.000 claims abstract description 9
- 238000004458 analytical method Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 239000012634 fragment Substances 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 238000006073 displacement reaction Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 9
- 230000006870 function Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 101100481876 Danio rerio pbk gene Proteins 0.000 description 2
- 101100481878 Mus musculus Pbk gene Proteins 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000011425 standardization method Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- 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
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
-
- 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
- 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
-
- 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)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a quick search method and a quick search system for optimal cascade configuration of video target detection, wherein the method comprises the following steps: acquiring a video data set and performing feature calculation on the video data set to obtain a scene feature combination; performing a frame filtering operation on the video data set based on the inter-frame similarity; based on an optimized configuration search algorithm, acquiring an optimal cascade configuration scheme which meets the precision requirement and has the lowest cost in a video data set, and constructing to obtain a training set with labels; training a cascade scheme mapper based on a training set to obtain a trained cascade scheme mapper; and acquiring a video to be detected, searching for the optimal configuration based on the trained cascade scheme mapper, and completing a target detection task. The system comprises: the device comprises a feature calculation module, a filtering module, a searching module, a training module and a detection module. By using the method and the device, the optimal cascade scheme can be automatically and efficiently obtained according to the video scene, and the target detection is completed. The invention can be widely applied to the field of target detection.
Description
Technical Field
The invention relates to the field of target detection, in particular to a quick search method and a quick search system for optimal cascade configuration of video target detection.
Background
At present, with the development of computer vision technology, the target detection model based on the deep neural network has increasingly accurate results. However, it presents a significant challenge to use for large-scale video data sets. Mainly because the video target detection task is executed by mobile equipment such as a camera, and the high calculation cost cannot be borne. Therefore, it is important to research how to reduce the calculation cost under the requirement of meeting the detection precision.
Disclosure of Invention
In order to solve the above technical problems, an object of the present invention is to provide a method and a system for quickly searching for an optimal cascade configuration for video target detection, which can automatically and efficiently obtain an optimal cascade scheme according to a video scene to complete target detection.
The first technical scheme adopted by the invention is as follows: a fast search method for video target detection optimal cascade configuration comprises the following steps:
acquiring a video data set and performing feature calculation on the video data set to obtain a scene feature combination;
performing frame filtering operation on the video data set based on the inter-frame similarity to obtain a video data set with redundant frames filtered out;
based on an optimized configuration search algorithm, acquiring an optimal cascade configuration scheme which meets the precision requirement and has the lowest cost in a video data set, and combining scene characteristic combination to construct a training set with labels;
training a cascade scheme mapper based on a training set to obtain a trained cascade scheme mapper;
and acquiring a video to be detected, searching for the optimal configuration based on the trained cascade scheme mapper, and completing a target detection task.
Further, before the step of obtaining the video data set and performing feature calculation on the video data set to obtain the scene feature value, the method further includes:
and (5) counting the running cost of all the configuration schemes.
Further, the step of obtaining a video data set and performing feature calculation on the video data set to obtain a scene feature combination specifically includes:
acquiring a video data set, wherein the video data set comprises a detection data set, a network video and a shooting video;
carrying out scene analysis on the video data set, calculating the characteristics of each frame of picture in the video data set, preprocessing the characteristics, and extracting to obtain a scene characteristic combination;
the scene feature combination comprises the number of detection targets, the speed of the detection targets, the displacement of the detection targets, the scene offset and the CNN feature.
Further, the preprocessing comprises normalization processing and extreme outlier removal processing by a 0-1 normalization method.
Further, the step of performing a frame filtering operation on the video data set based on the inter-frame similarity to obtain a video data set with redundant frames filtered out specifically includes:
acquiring an interframe difference algorithm;
comparing the calculation cost of the interframe difference algorithm, the adaptability of the scene and the filtering threshold interval;
and selecting an interframe difference algorithm to calculate interframe similarity, and filtering the video frames to obtain filtered video frame data.
Further, the optimal cascade configuration scheme which meets the precision requirement and has the lowest cost in the video data set is obtained based on the optimized configuration search algorithm, and the step of constructing the training set with the labels by combining the scene features specifically comprises the following steps:
dividing videos in the video data set into preset lengths in sequence to obtain a segment set;
carrying out filtering strategy and cascade configuration analysis on each fragment in the fragment set to obtain a corresponding combination scheme;
and generating video data with labels according to the scene feature combination and the combination scheme corresponding to the video data set, and constructing to obtain a training set with labels.
Further, the step of training the mapper for the cascading scheme based on the training set to obtain the mapper for the trained cascading scheme includes:
based on the training set, taking scene feature combination as input, and taking a combination scheme as an output training cascade scheme mapper;
and drawing an accuracy rate change curve in the training process, and debugging the cascade scheme mapper until the accuracy rate is judged to reach a preset value, so as to obtain the cascade scheme mapper after training.
Further, the step of obtaining the video to be detected, searching for the optimal configuration based on the trained cascade scheme mapper, and completing the target detection task specifically includes:
constructing and operating a trained cascade scheme mapper on the NCNN;
inputting a video to be detected on line, and performing feature extraction processing on the video to be detected to obtain the features of a scene to be detected;
outputting an optimal combination according to the characteristics of the scene to be detected, wherein the optimal combination comprises cascade configuration and a filtering strategy;
and filtering the video to be detected according to the optimal combination, and finishing a video target detection task by combining cascade configuration.
The second technical scheme adopted by the invention is as follows: a fast search system for optimal cascade configuration of video object detection, comprising:
the characteristic calculation module is used for acquiring a video data set and performing characteristic calculation on the video data set to obtain a scene characteristic combination;
the filtering module executes frame filtering operation on the video data set based on the interframe similarity to obtain the video data set with redundant frames filtered out;
the search module is used for acquiring an optimal cascade configuration scheme which meets the precision requirement and has the lowest cost in the video data set based on an optimized configuration search algorithm, and combining scene characteristics to construct a training set with labels;
the training module is used for training the cascade scheme mapper based on the training set to obtain a trained cascade scheme mapper;
and the detection module is used for acquiring the video to be detected, searching the optimal configuration based on the trained cascade scheme mapper and completing the target detection task.
The method and the system have the beneficial effects that: the invention can efficiently select effective filtering strategies and cascade configuration according to scene characteristics. After the frames are filtered, the lightweight configuration is preferentially considered from the cascade configuration of the current scene, and if the requirements are not met, the high-precision configuration is considered, so that the high efficiency, high precision and lowest cost of the whole target detection process are realized.
Drawings
FIG. 1 is a flow chart of the steps of a fast search method for optimal cascade configuration of video target detection according to the present invention;
fig. 2 is a block diagram of a fast search system of an optimal cascade configuration for video object detection according to the present invention.
Detailed Description
The invention is described in further detail below with reference to the figures and the specific embodiments. The step numbers in the following embodiments are provided only for convenience of illustration, the order between the steps is not limited at all, and the execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.
As shown in fig. 1, the present invention provides a fast search method for detecting an optimal cascade configuration of a video target, which includes the following steps:
and S0, counting the running cost of all the configuration schemes.
S1, acquiring a video data set and performing feature calculation on the video data set to obtain a scene feature combination;
s1.1, acquiring a video data set, wherein the video data set comprises a detection data set, a network video and a shooting video;
specifically, data covering a large number of common target detection task scenes can be acquired through channels such as downloading mainstream target detection data sets (such as KITTI, VOC, COCO and the like), network videos or self-shooting videos.
S1.2, performing scene analysis on the video data set, calculating the characteristics of each frame of picture in the video data set, preprocessing the characteristics, and extracting to obtain a scene characteristic combination;
the scene feature combination comprises the number of detection targets, the speed of the detection targets, the displacement of the detection targets, the scene offset and CNN features, and the preprocessing comprises the steps of carrying out normalization processing and extreme outlier removal processing by adopting a 0-1 standardization method.
Specifically, the normalized data conforms to the standard normal distribution, that is, the mean value is 0 and the standard deviation is 1, so that the dimensional influence between different characteristic values can be effectively removed.
S2, performing frame filtering operation on the video data set based on the interframe similarity to obtain the video data set with redundant frames filtered out;
s2.1, acquiring an interframe difference algorithm;
s2.2, comparing the calculation cost of the interframe difference algorithm, the adaptability of the scene and a filtering threshold interval;
specifically, a commonly used calculation method for measuring inter-frame difference in an industrial frame filtering strategy, such as Mean Square Error (MSE), Mean Absolute Error (MAE), peak signal to noise ratio (PSNR), Structural Similarity (SSIM), and the like, is investigated, the calculation costs of different difference algorithms and the adaptability to a scene are compared, and filtering threshold intervals of different inter-frame difference calculation methods are determined.
And S2.3, selecting an interframe difference algorithm to filter the video frames to obtain filtered video frame data.
S3, obtaining an optimal cascade configuration scheme which meets the precision requirement and has the lowest cost in the video data set based on an optimized configuration search algorithm, and combining scene characteristics to construct a training set with labels;
specifically, because the subsequent training neural network belongs to supervised learning, and the precondition is that high-quality labeled data are obtained, an optimized searching mode is adopted, the time complexity of the searching cascade space is reduced from exponential o (m ^ n) to o (m ^ n), the optimal cascade scheme under a large number of scenes is efficiently obtained, and the high-quality labeled training data are generated for the subsequent mapper based on the neural network training cascade scheme.
S3.1, sequentially dividing videos in the video data set into preset lengths to obtain a segment set;
specifically, video data is sequentially divided into a set of segments of length 16s (referred to as Window). Each Window is in turn divided into small segments (called segments) of length 4 s.
S3.2, performing filtering strategy and cascade configuration analysis on each fragment in the fragment set to obtain a corresponding combination scheme;
specifically, a first Segment in each Window is subjected to comprehensive filtering strategy and cascade configuration analysis, an optimal combination scheme is obtained, and the combination of top K filtering strategy and cascade configuration is recorded (namely, the K group scheme which meets the precision requirement and has the minimum calculation cost, namely filtering strategy and secondary cascade configuration).
Aiming at each remaining Segemt belonging to the same Window, the comprehensive analysis of the combination scheme is not executed, the optimal cascade configuration and the filtering strategy are only separated out from the top K scheme, the combination search space is greatly reduced, and the optimal filtering strategy and the cascade secondary configuration of the Segment are obtained.
If the optimal combination satisfying the accuracy requirement cannot be obtained from the topK combination propagated from the Segment, the scene is changed greatly, and at this time, the topK combination needs to be re-analyzed and updated.
And S3.3, generating video data with labels according to the scene feature combination and the combination scheme corresponding to the video data set, and constructing to obtain a training set.
S4, training the cascade scheme mapper based on the training set to obtain the trained cascade scheme mapper;
s4.1, based on a training set, combining scene features as input, and taking a combination scheme as an output training cascade scheme mapper;
in addition, test sets may also be constructed for verification.
And S4.2, drawing an accuracy rate change curve in the training process, and debugging the cascade scheme mapper until the accuracy rate is judged to reach a preset value, so as to obtain the trained cascade scheme mapper.
Specifically, the debugging operations are: changing the learning rate, observing a convergence curve of the target function value and the accuracy rate of the verification set, and selecting a proper learning rate from the convergence curve; changing the number of hidden layers and the number of hidden units in each layer, observing the change of accuracy rate in each change process, and selecting proper number of layers and units; and (5) observing a training set accuracy rate change curve and a verification set accuracy rate change curve while changing the number of training rounds, and recording the most appropriate number of training rounds.
And S5, acquiring the video to be detected, searching for the optimal configuration based on the trained cascade scheme mapper, and completing the target detection task.
S5.1, building and operating the trained cascade scheme mapper on the NCNN;
particularly, the NCNN is a high-performance neural network forward computing framework extremely optimized for mobile equipment, and can efficiently complete a video target detection task. The method can use a target detection model based on the deep neural network on a large-scale video data set on a movable device with limited computing resources, and reduces computing overhead.
S5.2, inputting a video to be detected on line, and performing feature extraction processing on the video to be detected to obtain the scene features to be detected;
s5.3, outputting an optimal combination according to the characteristics of the scene to be detected, wherein the optimal combination comprises cascade configuration and a filtering strategy;
and S5.4, filtering the video to be detected according to the optimal combination, and finishing a video target detection task by combining cascade configuration.
Specifically, video data is input, and extracted features which can describe scene change conditions are calculated. Since the selected features are key visual features, they can be calculated quickly. Inputting the characteristic value into a combined mapper to obtain output: the best combination, i.e. the best cascade configuration and the best filtering strategy. According to the selected optimal combination, a corresponding frame filtering strategy is executed, and frames can be directly filtered with low difference degree (the result of the previous frame is multiplexed, and model scheduling is avoided). Then, the best configuration is obtained from the cascaded two-stage configuration: the lightweight configuration is considered first, and the heavyweight configuration is considered only when the precision does not reach the standard. The optimal configuration is then used, according to the parameter values: frame rate, resolution and object detection model, completing the video object detection task.
The invention optimizes the mainstream video target detection process based on the neural network. The method can adapt to scene change, periodically adjust the optimal cascade combination scheme for completing target detection, and ensure the balance of precision and calculation cost. The method mainly comprises the steps of analyzing scene change, calculating scene characteristics, inputting the characteristics to a cascade combination mapper based on neural network training, directly outputting the optimal cascade combination, and efficiently selecting the cascade combination with the lowest cost meeting the precision requirement from a huge cascade combination space. The frame filtering model automatically selects a proper filtering strategy and a threshold value according to the scene, and can dynamically adjust the filtering decision according to the scene characteristic type, the query precision and the time-varying correlation among the video contents. In addition, the selected method for measuring the inter-frame difference is high in calculation speed, and the frame filtering model does not bring high overhead. The cascade configuration will also be dynamically adjusted according to the scene.
In the off-line stage, the optimized search algorithm is used, so that the cascade combination search space is greatly reduced, and high-quality labeled data can be efficiently obtained by training the cascade combination mapper by utilizing the neural network. Due to the adoption of the optimized searching scheme, the searching speed is high. This step is performed off-line, preventing the cost of analyzing the optimal cascade combination from offsetting the gains from switching the cascade combination. In addition, by building a neural network, the relation between the scene and the optimal cascade combination is learned, and the method can be adapted to most scenes in real life. In the online stage, the rapid search of the optimal cascade combination of the video target detection is completed by inputting the scene characteristic information, so that the prediction precision is improved, the resource consumption of the current video detection is obviously reduced, and the effects of low cost, high efficiency and high accuracy are achieved.
As shown in fig. 2, a fast search system for an optimal cascade configuration of video object detection includes:
the characteristic calculation module is used for acquiring a video data set and performing characteristic calculation on the video data set to obtain a scene characteristic combination;
the filtering module executes frame filtering operation on the video data set based on the interframe similarity to obtain the video data set with redundant frames filtered out;
the search module is used for acquiring an optimal cascade configuration scheme which meets the precision requirement and has the lowest cost in the video data set based on an optimized configuration search algorithm, and combining scene characteristics to construct a training set with labels;
the training module is used for training the cascade scheme mapper based on the training set to obtain a trained cascade scheme mapper;
and the detection module is used for acquiring the video to be detected, searching the optimal configuration based on the trained cascade scheme mapper and completing the target detection task.
Further as a preferred embodiment, the method further comprises:
and the pre-calculation module is used for counting the running cost of all the configuration schemes.
The contents in the above method embodiments are all applicable to the present system embodiment, the functions specifically implemented by the present system embodiment are the same as those in the above method embodiment, and the beneficial effects achieved by the present system embodiment are also the same as those achieved by the above method embodiment.
A fast search device for video target detection optimal cascade configuration comprises:
at least one processor;
at least one memory for storing at least one program;
when executed by the at least one processor, the at least one program causes the at least one processor to implement a fast search method for detecting an optimal cascade configuration of video objects as described above.
The contents in the above method embodiments are all applicable to the present apparatus embodiment, the functions specifically implemented by the present apparatus embodiment are the same as those in the above method embodiments, and the advantageous effects achieved by the present apparatus embodiment are also the same as those achieved by the above method embodiments.
A storage medium having stored therein instructions executable by a processor, the storage medium comprising: the processor-executable instructions, when executed by the processor, are for implementing a fast search method for video object detection optimal concatenation configuration as described above.
The contents in the above method embodiments are all applicable to the present storage medium embodiment, the functions specifically implemented by the present storage medium embodiment are the same as those in the above method embodiments, and the advantageous effects achieved by the present storage medium embodiment are also the same as those achieved by the above method embodiments.
While the invention has been described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (9)
1. A fast search method for video target detection optimal cascade configuration is characterized by comprising the following steps:
acquiring a video data set and performing feature calculation on the video data set to obtain a scene feature combination;
performing frame filtering operation on the video data set based on the inter-frame similarity to obtain a video data set with redundant frames filtered out;
based on an optimized configuration search algorithm, acquiring an optimal cascade configuration scheme which meets the precision requirement and has the lowest cost in a video data set, and combining scene characteristic combination to construct a training set with labels;
training a cascade scheme mapper based on a training set to obtain a trained cascade scheme mapper;
and acquiring a video to be detected, searching for the optimal configuration based on the trained cascade scheme mapper, and completing a target detection task.
2. The method of claim 1, wherein before the steps of obtaining the video data set and performing feature calculation on the video data set to obtain the scene feature combination, the method further comprises:
and (5) counting the running cost of all the configuration schemes.
3. The method according to claim 2, wherein the step of obtaining the video data set and performing feature calculation on the video data set to obtain the scene feature combination specifically comprises:
acquiring a video data set, wherein the video data set comprises a detection data set, a network video and a shooting video;
carrying out scene analysis on the video data set, calculating the characteristics of each frame of picture in the video data set, preprocessing the characteristics, and extracting to obtain a scene characteristic combination;
the scene feature combination comprises the number of detection targets, the speed of the detection targets, the displacement of the detection targets, the scene offset and the CNN feature.
4. The fast search method for optimal cascade configuration of video object detection as claimed in claim 3, wherein the preprocessing comprises normalization and outlier removal processing using 0-1 normalization.
5. The method of claim 4, wherein the step of performing a frame filtering operation on the video data set based on the inter-frame similarity to obtain a video data set with redundant frames filtered includes:
acquiring an interframe difference algorithm;
comparing the calculation cost of the interframe difference algorithm, the adaptability of the scene and the filtering threshold interval;
and selecting an interframe difference algorithm to calculate interframe similarity, and filtering the video frames to obtain filtered video frame data.
6. The method according to claim 5, wherein the step of obtaining the optimal cascade configuration scheme with the lowest cost and meeting the accuracy requirement in the video data set based on the optimized configuration search algorithm and constructing the training set with the tags by combining the scene features specifically comprises:
dividing videos in the video data set into preset lengths in sequence to obtain a segment set;
carrying out filtering strategy and cascade configuration analysis on each fragment in the fragment set to obtain a corresponding combination scheme;
and generating video data with labels according to the scene feature combination and combination scheme corresponding to the video data set, and constructing to obtain a training set with labels.
7. The method as claimed in claim 6, wherein the step of training the mapper based on the training set to obtain a trained mapper comprises:
based on the training set, the scene feature combination is used as input, and the combination scheme is used as output training cascade scheme mapper;
and drawing an accuracy rate change curve in the training process, debugging the cascade scheme mapper, and judging that the accuracy rate reaches a preset value to obtain the trained cascade scheme mapper.
8. The method as claimed in claim 7, wherein the step of obtaining the video to be detected and searching for the optimal configuration based on the trained mapper of the cascading scheme to complete the target detection task includes:
constructing and operating a trained cascade scheme mapper on the NCNN;
inputting a video to be detected on line, and performing feature extraction processing on the video to be detected to obtain the features of a scene to be detected;
outputting an optimal combination according to the characteristics of the scene to be detected, wherein the optimal combination comprises cascade configuration and a filtering strategy;
and filtering the video to be detected according to the optimal combination, and finishing a video target detection task by combining cascade configuration.
9. A fast search system for optimal cascading configuration of video object detection, comprising:
the characteristic calculation module is used for acquiring a video data set and performing characteristic calculation on the video data set to obtain a scene characteristic combination;
the filtering module executes frame filtering operation on the video data set based on the interframe similarity to obtain the video data set with redundant frames filtered out;
the searching module is used for acquiring an optimal cascade configuration scheme which meets the precision requirement and has the lowest cost in the video data set based on an optimized configuration searching algorithm, and combining scene characteristics to construct a training set with labels;
the training module is used for training the cascade scheme mapper based on the training set to obtain a trained cascade scheme mapper;
and the detection module is used for acquiring the video to be detected, searching the optimal configuration based on the trained cascade scheme mapper and completing the target detection task.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210147889.0A CN114550036B (en) | 2022-02-17 | 2022-02-17 | Searching method and system for optimal cascade configuration of video target detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210147889.0A CN114550036B (en) | 2022-02-17 | 2022-02-17 | Searching method and system for optimal cascade configuration of video target detection |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114550036A true CN114550036A (en) | 2022-05-27 |
| CN114550036B CN114550036B (en) | 2024-08-16 |
Family
ID=81675162
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210147889.0A Active CN114550036B (en) | 2022-02-17 | 2022-02-17 | Searching method and system for optimal cascade configuration of video target detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114550036B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110991658A (en) * | 2019-11-28 | 2020-04-10 | 重庆紫光华山智安科技有限公司 | Model training method and device, electronic equipment and computer readable storage medium |
| CN112418055A (en) * | 2020-11-18 | 2021-02-26 | 东方通信股份有限公司 | Scheduling method based on video analysis and personnel trajectory tracking method |
| WO2021055442A1 (en) * | 2019-09-18 | 2021-03-25 | Google Llc | Small and fast video processing networks via neural architecture search |
| CN113283426A (en) * | 2021-04-30 | 2021-08-20 | 南京大学 | Embedded target detection model generation method based on multi-target neural network search |
| CN113869521A (en) * | 2020-06-30 | 2021-12-31 | 华为技术有限公司 | Method, device, computing equipment and storage medium for constructing prediction model |
-
2022
- 2022-02-17 CN CN202210147889.0A patent/CN114550036B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021055442A1 (en) * | 2019-09-18 | 2021-03-25 | Google Llc | Small and fast video processing networks via neural architecture search |
| CN110991658A (en) * | 2019-11-28 | 2020-04-10 | 重庆紫光华山智安科技有限公司 | Model training method and device, electronic equipment and computer readable storage medium |
| CN113869521A (en) * | 2020-06-30 | 2021-12-31 | 华为技术有限公司 | Method, device, computing equipment and storage medium for constructing prediction model |
| CN112418055A (en) * | 2020-11-18 | 2021-02-26 | 东方通信股份有限公司 | Scheduling method based on video analysis and personnel trajectory tracking method |
| CN113283426A (en) * | 2021-04-30 | 2021-08-20 | 南京大学 | Embedded target detection model generation method based on multi-target neural network search |
Non-Patent Citations (2)
| Title |
|---|
| ZHAOWEI CAI: "Cascade R-CNN High Quality Object Detection and Instance Segmentation", 《ARXIV》, 24 June 2019 (2019-06-24) * |
| 谭光等: "基于多方向搜索的连续帧目标检测去重算法", 《华中科技大学学报》, 30 November 2021 (2021-11-30) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114550036B (en) | 2024-08-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109711228B (en) | Image processing method and device for realizing image recognition and electronic equipment | |
| CN111968150B (en) | Weak surveillance video target segmentation method based on full convolution neural network | |
| Liu et al. | Learning task-specific representation for video anomaly detection with spatial-temporal attention | |
| CN110222592B (en) | Construction method of time sequence behavior detection network model based on complementary time sequence behavior proposal generation | |
| CN111831852B (en) | Video retrieval method, device, equipment and storage medium | |
| CN110674886B (en) | Video target detection method fusing multi-level features | |
| CN115587335A (en) | Training method of abnormal value detection model, abnormal value detection method and system | |
| CN116152722B (en) | Video anomaly detection method based on combination of residual attention block and self-selection learning | |
| CN112200096A (en) | Method, device and storage medium for realizing real-time abnormal behavior recognition based on compressed video | |
| CN118396576B (en) | Construction progress monitoring and early warning method and system | |
| CN110348509B (en) | Method, device and equipment for adjusting data augmentation parameters and storage medium | |
| CN114550036B (en) | Searching method and system for optimal cascade configuration of video target detection | |
| CN116934796B (en) | Visual target tracking method based on twinning residual error attention aggregation network | |
| CN117218382A (en) | Unmanned system large-span shuttle multi-camera track tracking and identifying method | |
| CN114819110B (en) | Method and device for identifying speaker in video in real time | |
| CN112487927B (en) | Method and system for realizing indoor scene recognition based on object associated attention | |
| CN112767438B (en) | Multi-target tracking method combining space-time motion | |
| CN109873987B (en) | Target searching method and system based on monitoring video | |
| CN112052626A (en) | Automatic neural network design system and method | |
| CN111783655A (en) | Image processing method and device, electronic equipment and storage medium | |
| CN117409354B (en) | Video anomaly detection method and system based on three paths of video streams and context awareness | |
| CN116170638B (en) | Self-attention video stream compression method and system for online action detection task | |
| CN111476131A (en) | Video processing method and device | |
| Diao et al. | A slowfast-based violence recognition method | |
| CN111291602A (en) | Video detection method and device, electronic equipment and computer readable storage medium |
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 |