CN113837022A - Method for rapidly searching video pedestrian - Google Patents
Method for rapidly searching video pedestrian Download PDFInfo
- Publication number
- CN113837022A CN113837022A CN202111025272.3A CN202111025272A CN113837022A CN 113837022 A CN113837022 A CN 113837022A CN 202111025272 A CN202111025272 A CN 202111025272A CN 113837022 A CN113837022 A CN 113837022A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- similarity
- target
- video
- similar
- 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
- 238000000034 method Methods 0.000 title claims abstract description 26
- 239000013598 vector Substances 0.000 claims abstract description 44
- 238000001514 detection method Methods 0.000 claims abstract description 4
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000005259 measurement Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 abstract description 3
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000012805 post-processing Methods 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
- G06F18/22—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
- G06F18/23—Clustering techniques
Abstract
The invention discloses a method for quickly searching video pedestrians, which comprises the following steps: s1, selecting a target picture to be retrieved and simultaneously acquiring a video to be retrieved; s2, acquiring all pedestrian target images of each frame in the video through target detection, forming a pedestrian image set according to the pedestrian target images, and giving IDs to the pedestrian target images; s3, extracting feature vectors of the target picture and pictures in the pedestrian image set; s4, clustering the characteristic vectors of the pedestrian images of the video to obtain at least one cluster, and selecting a corresponding cluster center; and S5, solving the similarity between the target pedestrian feature vector and the clustering center, and solving the similarity between the target pedestrian feature vector and all feature vectors in the similarity class to obtain the similar pedestrian. The invention greatly reduces the calculation cost, saves the calculated amount for large-scale pedestrian retrieval and improves the time efficiency.
Description
Technical Field
The invention relates to the field of image retrieval, in particular to a method for quickly searching video pedestrians.
Background
Along with the construction of a smart city, the monitoring video of the road is widely applied, and the public security of the city accumulates massive video data. Searching a target suspect in a large amount of videos by using an artificial intelligence technology is an important work at present, but when the amount of videos to be searched is too large, the work has the problem of insufficient computing resources. Because the video is continuous, 25 frames of images exist in the video of 1 second in the common camera, when the video is searched for more than 10 hours, the pedestrian images in each frame of image need to be extracted first, the number is huge, the traditional mode is violent search, the images to be compared are compared one by one, the complexity of the working time and the complexity of the space are high, and a large amount of GPU computing resources and computing time are consumed.
Patent CN201910606483.2 discloses a video retrieval method and system, in which a target object appearing in a video is labeled in advance, the label of the target object to be searched is input during post-processing, and the target object appearing in the video is retrieved by means of label matching. The method can also be used for searching the pedestrians in the video, the pedestrians are labeled according to the appearance, and the label search is carried out afterwards, but the method cannot identify the pedestrians with the transshipment behaviors through the labels, so the method is invalid in some scenes.
A method for rapidly searching for video pedestrians is needed to solve the above problems.
Disclosure of Invention
The invention aims to solve the problems that in the prior art, target pedestrian feature vectors are compared with all pedestrian feature vectors one by one, the method is more efficient when videos of a small number of pedestrians are processed, but when high-density crowds appear in the videos, the problem of video memory overflow during calculation is caused, the method for quickly searching the video pedestrians is provided, the pedestrian feature vectors are clustered, the clustering center is used as an index value, the similarity between the pedestrian feature vectors and the target feature vectors is calculated, and the similar image searching range is narrowed, so that the problems are solved.
The invention provides a method for quickly searching video pedestrians, which comprises the following steps:
s1, selecting a target picture to be retrieved and simultaneously acquiring a video to be retrieved;
s2, acquiring all pedestrian target images of each frame in the video through target detection, forming a pedestrian image set according to the pedestrian target images, and giving IDs to the pedestrian target images;
s3, extracting feature vectors of the pictures in the target picture and pedestrian image set;
s4, clustering the characteristic vectors of the pedestrian images of the video to obtain at least one cluster, and selecting a corresponding cluster center;
and S5, solving the similarity between the target pedestrian feature vector and the clustering center, and solving the similarity between the target pedestrian feature vector and all feature vectors in the similarity class to obtain the similar pedestrian.
As a preferred mode, the step S4 of the method for quickly searching video pedestrians in the present invention specifically includes:
s41, clustering the pedestrian features of the initial magnitude into n classes through distance measurement;
and S42, establishing a data index according to the clusters, wherein the cluster center of the cluster corresponding to each category is used as an index value.
As a preferred mode, the step S5 of the method for quickly searching video pedestrians in the present invention specifically includes:
s51, setting a similarity threshold value a;
s52, solving a first similarity between the target pedestrian feature vector and the n index values;
s53, judging whether the similarity is larger than a one by one, if so, taking the class corresponding to the first similarity as a similar class, and if not, giving up the class corresponding to the current similarity;
s54, merging all the eigenvectors in the similar classes to form an eigenvector matrix;
s55, solving a second similarity between the feature vector of the target pedestrian and the similar feature matrix;
s56, judging whether the second similarity is larger than a, if so, taking the current pedestrian ID as a similar crowd, otherwise, giving up the current pedestrian ID;
and S57, storing data of pedestrians of similar crowds in the video, comparing the characteristic vector of the pedestrian with the index values in the database one by one, and putting the pedestrian vector into the most similar index class.
The invention relates to a method for rapidly searching video pedestrians, which is a preferred mode, wherein a first similarity formula is as follows:
wherein COS θ1Is the first cosine similarity, and the second cosine similarity,
is the characteristic vector of the target pedestrian,
is a vector of index values.
The invention relates to a method for rapidly searching video pedestrians, which is a preferred mode, wherein a second similarity formula is as follows:
wherein COS θ2Is the second cosine similarity, and is,
is the characteristic vector of the target pedestrian,
is a similar feature matrix.
The invention has the following beneficial effects:
the pedestrian feature vectors are clustered, the clustering center is used as an index value, the similarity with the target feature vector is calculated to reduce the similar image searching range, the calculation cost is greatly reduced, the calculated amount is saved for large-scale pedestrian retrieval, and the time efficiency is improved.
Drawings
Fig. 1 is a flow chart of a method for video pedestrian fast search.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments.
Example 1
As shown in fig. 1, a method for video pedestrian fast search includes the following steps:
s1, selecting a target picture to be retrieved and simultaneously acquiring a video to be retrieved;
s2, acquiring all pedestrian target images of each frame in the video through target detection, forming a pedestrian image set according to the pedestrian target images, and giving IDs to the pedestrian target images;
s3, extracting feature vectors of the pictures in the target picture and pedestrian image set;
s4, clustering the pedestrian features of the initial magnitude into n classes through distance measurement;
s5, establishing a data index according to the clusters, wherein the cluster center of the cluster corresponding to each category is used as an index value;
s6, setting a similarity threshold value a;
s7, solving a first similarity between the target pedestrian feature vector and the n index values;
s8, judging whether the similarity is larger than a one by one, if so, taking the class corresponding to the first similarity as a similar class, and if not, giving up the class corresponding to the current similarity;
s9, merging all the eigenvectors in the similar classes to form an eigenvector matrix;
s10, solving a second similarity between the feature vector of the target pedestrian and the similar feature matrix;
s11, judging whether the second similarity is larger than a, if so, taking the current pedestrian ID as a similar crowd, otherwise, giving up the current pedestrian ID;
and S12, storing data of pedestrians of similar crowds in the video, comparing the characteristic vector of the pedestrian with the index values in the database one by one, and putting the pedestrian vector into the most similar index class.
The first similarity formula is:
wherein COS θ1Is the first cosine similarity，
Is the characteristic vector of the target pedestrian,
is a vector of index values.
The second similarity formula is:
wherein COS θ2Is the second cosine similarity, and is,
is the characteristic vector of the target pedestrian,
is a similar feature matrix.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.
Claims (5)
1. A method for quickly searching video pedestrians is characterized by comprising the following steps: the method comprises the following steps:
s1, selecting a target picture to be retrieved and simultaneously acquiring a video to be retrieved;
s2, acquiring all pedestrian target images of each frame in the video through target detection, forming a pedestrian image set according to the pedestrian target images, and giving IDs to the pedestrian target images;
s3, extracting feature vectors of the target picture and pictures in the pedestrian image set;
s4, clustering the characteristic vectors of the pedestrian images of the video to obtain at least one cluster, and selecting a corresponding cluster center;
and S5, solving the similarity between the target pedestrian feature vector and the clustering center, and solving the similarity between the target pedestrian feature vector and all feature vectors in the similarity class to obtain the similar pedestrian.
2. The method for video pedestrian fast search according to claim 1, characterized in that: the step S4 specifically includes:
s41, clustering the pedestrian features of the initial magnitude into n classes through distance measurement;
and S42, establishing a data index according to the clusters, wherein the cluster center of the cluster corresponding to each category is used as an index value.
3. The method for video pedestrian fast search according to claim 2, characterized in that: the step S5 specifically includes:
s51, setting a similarity threshold value a;
s52, solving a first similarity between the target pedestrian feature vector and the n index values;
s53, judging whether the similarity is larger than a one by one, if so, taking the class corresponding to the first similarity as a similar class, and if not, giving up the class corresponding to the current similarity;
s54, merging all the eigenvectors in the similar classes to form an eigenvector matrix;
s55, solving a second similarity between the feature vector of the target pedestrian and the similar feature matrix;
s56, judging whether the second similarity is larger than a, if so, taking the current pedestrian ID as a similar crowd, otherwise, giving up the current pedestrian ID;
and S57, storing data of pedestrians of similar crowds in the video, comparing the characteristic vector of the pedestrian with the index values in the database one by one, and putting the pedestrian vector into the most similar index class.
4. The method of claim 3, wherein the pedestrian video fast search method comprises: the first similarity formula is as follows:
wherein COS θ1Is the first cosine similarity, and the second cosine similarity,
is the characteristic vector of the target pedestrian,
is a vector of index values.
5. The method of claim 3, wherein the pedestrian video fast search method comprises: the second similarity formula is as follows:
wherein COS θ2Is the second cosine similarity, and is,
is the characteristic vector of the target pedestrian,
is a similar feature matrix.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111025272.3A CN113837022A (en) | 2021-09-02 | 2021-09-02 | Method for rapidly searching video pedestrian |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111025272.3A CN113837022A (en) | 2021-09-02 | 2021-09-02 | Method for rapidly searching video pedestrian |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113837022A true CN113837022A (en) | 2021-12-24 |
Family
ID=78962071
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111025272.3A Pending CN113837022A (en) | 2021-09-02 | 2021-09-02 | Method for rapidly searching video pedestrian |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113837022A (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105760472A (en) * | 2016-02-06 | 2016-07-13 | 中国农业大学 | Video retrieval method and system |
| CN106778501A (en) * | 2016-11-21 | 2017-05-31 | 武汉科技大学 | Video human face ONLINE RECOGNITION method based on compression tracking with IHDR incremental learnings |
| CN110309362A (en) * | 2019-07-05 | 2019-10-08 | 深圳中科云海科技有限公司 | A kind of video retrieval method and system |
| CN110414462A (en) * | 2019-08-02 | 2019-11-05 | 中科人工智能创新技术研究院(青岛)有限公司 | A kind of unsupervised cross-domain pedestrian recognition methods and system again |
| CN110609916A (en) * | 2019-09-25 | 2019-12-24 | 四川东方网力科技有限公司 | Video image data retrieval method, device, equipment and storage medium |
| CN110659576A (en) * | 2019-08-23 | 2020-01-07 | 深圳久凌软件技术有限公司 | Pedestrian searching method and device based on joint judgment and generation learning |
| CN110750671A (en) * | 2019-09-05 | 2020-02-04 | 浙江省北大信息技术高等研究院 | Pedestrian retrieval method and device based on massive unstructured features |
| CN111241345A (en) * | 2020-02-18 | 2020-06-05 | 腾讯科技（深圳）有限公司 | Video retrieval method and device, electronic equipment and storage medium |
| CN112633051A (en) * | 2020-09-11 | 2021-04-09 | 博云视觉（北京）科技有限公司 | Online face clustering method based on image search |
| CN112784102A (en) * | 2021-01-20 | 2021-05-11 | 北京百度网讯科技有限公司 | Video retrieval method and device and electronic equipment |
-
2021
- 2021-09-02 CN CN202111025272.3A patent/CN113837022A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105760472A (en) * | 2016-02-06 | 2016-07-13 | 中国农业大学 | Video retrieval method and system |
| CN106778501A (en) * | 2016-11-21 | 2017-05-31 | 武汉科技大学 | Video human face ONLINE RECOGNITION method based on compression tracking with IHDR incremental learnings |
| CN110309362A (en) * | 2019-07-05 | 2019-10-08 | 深圳中科云海科技有限公司 | A kind of video retrieval method and system |
| CN110414462A (en) * | 2019-08-02 | 2019-11-05 | 中科人工智能创新技术研究院(青岛)有限公司 | A kind of unsupervised cross-domain pedestrian recognition methods and system again |
| CN110659576A (en) * | 2019-08-23 | 2020-01-07 | 深圳久凌软件技术有限公司 | Pedestrian searching method and device based on joint judgment and generation learning |
| CN110750671A (en) * | 2019-09-05 | 2020-02-04 | 浙江省北大信息技术高等研究院 | Pedestrian retrieval method and device based on massive unstructured features |
| CN110609916A (en) * | 2019-09-25 | 2019-12-24 | 四川东方网力科技有限公司 | Video image data retrieval method, device, equipment and storage medium |
| CN111241345A (en) * | 2020-02-18 | 2020-06-05 | 腾讯科技（深圳）有限公司 | Video retrieval method and device, electronic equipment and storage medium |
| CN112633051A (en) * | 2020-09-11 | 2021-04-09 | 博云视觉（北京）科技有限公司 | Online face clustering method based on image search |
| CN112784102A (en) * | 2021-01-20 | 2021-05-11 | 北京百度网讯科技有限公司 | Video retrieval method and device and electronic equipment |
Non-Patent Citations (1)
| Title |
|---|
| 吴思霈: "云南旅游推荐系统信息提取与推荐算法研究", 《中国优秀硕士学位论文全文数据库 经济与管理科学辑》, 15 January 2018 (2018-01-15), pages 153 - 85 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109829443B (en) | Video behavior identification method based on image enhancement and 3D convolution neural network | |
| US10248860B2 (en) | System and method for object re-identification | |
| CN109961051B (en) | Pedestrian re-identification method based on clustering and block feature extraction | |
| Feris et al. | Large-scale vehicle detection, indexing, and search in urban surveillance videos | |
| WO2020098158A1 (en) | Pedestrian re-recognition method and apparatus, and computer readable storage medium | |
| WO2019169816A1 (en) | Deep neural network for fine recognition of vehicle attributes, and training method thereof | |
| CN102207966B (en) | Video content quick retrieving method based on object tag | |
| CN111639564B (en) | Video pedestrian re-identification method based on multi-attention heterogeneous network | |
| Yuan et al. | Traffic sign detection via graph-based ranking and segmentation algorithms | |
| WO2020206850A1 (en) | Image annotation method and device employing high-dimensional image | |
| Yu et al. | Motion pattern interpretation and detection for tracking moving vehicles in airborne video | |
| Li et al. | Effective person re-identification by self-attention model guided feature learning | |
| Xian et al. | Evaluation of low-level features for real-world surveillance event detection | |
| Molina-Moreno et al. | Efficient scale-adaptive license plate detection system | |
| CN110728216A (en) | Unsupervised pedestrian re-identification method based on pedestrian attribute adaptive learning | |
| CN103942563A (en) | Multi-mode pedestrian re-identification technology | |
| Aoun et al. | Graph modeling based video event detection | |
| Tseng et al. | Person retrieval in video surveillance using deep learning–based instance segmentation | |
| KR20200060868A (en) | multi-view monitoring system using object-oriented auto-tracking function | |
| AU2011265494A1 (en) | Kernalized contextual feature | |
| Yang et al. | Pedestrian detection in thermal infrared image using extreme learning machine | |
| KR101826669B1 (en) | System and method for video searching | |
| CN113837022A (en) | Method for rapidly searching video pedestrian | |
| Zhu et al. | Person re-identification in the real scene based on the deep learning | |
| Xiang et al. | Action recognition for videos by long-term point trajectory analysis with background removal |
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 |