A specific pedestrian target matching method based on multi-view feature crowd intelligence fusion
Through multi-view feature group intelligence fusion technology, multi-angle cameras are used to collect and process pedestrian image features, the problems of mismatch and mismatch caused by occlusion in video surveillance are solved, and higher matching accuracy and accuracy are achieved.
Patent Information
- Application Number
- CN202210768433.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-30
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2042-06-30
AI Technical Summary
In video surveillance, due to environmental factors such as occlusion of trees and buildings, the pedestrian target matching is difficult to effectively solve the problem of missing and mismatched.
Using a multi-view feature group intelligence fusion method, pedestrian images are collected through multi-angle cameras, SIFT features and spatial color histogram features are extracted, weighted processing and similarity calculation are performed, and mismatch probability and mismatch rate are reduced.
It effectively reduces the probability of mismatch caused by obstacles, improves the accuracy and accuracy of target matching for specific pedestrians, and reduces the mismatch rate.
Smart Images

Figure CN115100601B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of swarm intelligence, and in particular to a specific pedestrian target matching method based on multi-view feature swarm intelligence fusion. Background Art
[0002] In recent years, with the rapid development of video surveillance, group intelligent video surveillance will also be the development trend of video surveillance in the future, and pedestrian target matching technology is one of the important applications of intelligent security. Due to the occlusion of trees, buildings and other pedestrians in the monitoring scene, it is inevitable that the target pedestrian matching will be missed or mismatched. Summary of the invention
[0003] The purpose of the present invention is to provide a specific pedestrian target matching method based on multi-view feature group intelligence fusion to overcome the problem of missed matching and mismatching of the pedestrian to be matched and the target monitored pedestrian due to occlusion caused by environmental factors.
[0004] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a specific pedestrian target matching method based on multi-view feature group intelligence fusion, based on pedestrians in a target area and multi-angle cameras covering various positions in the target area, according to the preset angle images corresponding to each target monitored pedestrian in the target area, for each pedestrian to be matched in the target area, the following steps A to F are performed respectively to complete the primary matching of the person to be matched, and then step G is performed to achieve the matching of the pedestrian to be matched relative to the target monitored pedestrian;
[0005] Step A: for each target monitored pedestrian in the target area, extract the SIFT features and spatial color histogram feature information of each angle image corresponding to the target monitored pedestrian, and fuse them to obtain the multi-view fusion features of the target monitored pedestrian, that is, obtain the multi-view fusion features corresponding to each target monitored pedestrian, and construct the target monitored pedestrian feature set T; then execute step B;
[0006] Step B: for each angle image of the pedestrian to be matched, extract the view SIFT feature and the spatial color histogram feature information for each angle image, then fuse the view SIFT feature and the spatial color histogram feature information to obtain the view fusion features corresponding to each angle image of the pedestrian to be matched; then execute step C;
[0007] Step C: for each angle image of the pedestrian to be matched, the number of pixels of each angle image of the pedestrian to be matched is counted, and the image weight of each angle image is calculated to obtain the angle image with the largest weight, and then further obtain the camera corresponding to the angle image; and determine that the camera is the main camera in the target area; then execute step D;
[0008] Step D: Based on the image weights of each multi-angle image corresponding to the pedestrian to be matched and the multi-view fusion features of each multi-angle image, weighted processing is performed to obtain the weighted multi-view fusion features of the pedestrian to be matched under the main camera; then step E is performed;
[0009] Step E: Based on the monitored pedestrian feature set T and the weighted multi-view fusion features of the pedestrian to be matched under the main camera, the similarity between the monitored pedestrian feature set T and the weighted multi-view fusion features is calculated; further, the mismatch probability between the images of the pedestrian to be matched at each angle and the monitored pedestrian feature set T is obtained; and then step F is executed;
[0010] Step F: Determine the mismatch probability Q obtained in step E m Is it greater than or equal to the matching success threshold Q? th , if yes, the initial matching result is unsuccessful, and the process returns to step B; otherwise, the initial matching result is successful, and the person to be matched constitutes the primary person to be matched;
[0011] Step G: Calculate the similarity between each primary pedestrian to be matched and each target pedestrian to be monitored, and combine the preset similarity threshold β between the primary pedestrian to be matched and the target pedestrian to be monitored th , obtain the matching result between the primary pedestrian to be matched and the target monitored pedestrian.
[0012] Furthermore, in the aforementioned specific pedestrian target matching method based on multi-view feature group intelligence fusion, the multi-angle cameras covering various positions in the target area select k perspectives to shoot for each pedestrian, and obtain images of each pedestrian at various angles.
[0013] Furthermore, the aforementioned step A includes the following steps A1 to A3:
[0014] Step A1: Based on the images of each pedestrian at each angle, obtain the images of each monitored pedestrian at each angle I k , and images of pedestrians under surveillance from various angles I k The SIFT features of images at each angle are extracted respectively {P 1k ,…,P nk}, and the spatial color histogram features of images at each angle {Z 1k ,…,Z nk},in, Where n represents the number of monitored pedestrians, i∈[1,n];
[0015] Step A2: Based on the SIFT features of each monitored pedestrian angle image, {P 1k ,…,P nk}, and the spatial color histogram features of images at each angle {Z1k ,…,Z nk}, obtain the multi-view fusion features R of each monitored pedestrian i :
[0016] Among them, b represents the dimension of SIFT features,
[0017] Step A3: Based on the multi-view fusion features of each target monitored pedestrian, a target monitored pedestrian feature set T is constructed, where T = {R1,…,R n} and enter it into the server, where n represents the number of target monitored pedestrians.
[0018] Furthermore, the aforementioned step B includes the following steps B1 to B3:
[0019] Step B1: Extract the pedestrian t to be matched from the camera y perspective m SIFT features of images at different angles in represents the feature value on dimension s, b represents the dimension of SIFT feature, y represents the number of cameras arranged in the monitoring area, and one camera represents one viewing angle;
[0020] Step B2: Extract the pedestrian t to be matched m Spatial color histogram feature information of images at different angles in, represents the number of pixels whose absolute brightness is j;
[0021] Step B3: Treat the matched pedestrians t separately m The multi-view SIFT features P of each angle image my and spatial color histogram feature Z my Fusion is performed to obtain the pedestrian t to be matched m The perspective fusion features corresponding to the images at each angle are:
[0022] Wherein, b represents the dimension of SIFT features, y represents the number of cameras deployed in the monitoring area, and one camera represents one viewing angle.
[0023] Furthermore, the aforementioned step C includes the following steps C1 to C2;
[0024] Step C1: For each pedestrian t to be matched m Images of each angle, count each pedestrian t to be matched m The number of pixels N of each angle image y , 1≤y≤Y, Y represents the number of cameras deployed in the target domain;
[0025] Step C2: According to the following formula:
[0026]
[0027] Calculate each pedestrian t to be matched m The weight α of the angle image y ; and obtain the maximum weight α max ∈{α1,…,α Y} of the pedestrian t to be matched m , and further obtain the camera y corresponding to the pedestrian angle image to be matched with the largest weight. max ∈[1,Y], determine that the camera is the main camera.
[0028] Furthermore, the above step D is specifically as follows: based on each image of the pedestrian to be matched t m The angle image weight and the pedestrian t to be matched m Multi-view fusion features are fused according to the following formula:
[0029]
[0030] Obtain the weighted multi-view fusion features of the pedestrian to be matched under the main camera Where b represents the dimension of SIFT features, y represents the number of cameras deployed in the monitoring area, and one camera represents one viewing angle.
[0031] Furthermore, the aforementioned step E includes the following steps E1 to E2;
[0032] Step E1: Based on the pedestrian to be matched t m Weighted multi-view fusion features corresponding to the main camera And the multi-view fusion features R of each target monitored pedestrian in the monitored pedestrian feature set T i =(r i 1 ,…,r i b+256 ), i=1,…,n, according to the following formula:
[0033]
[0034] Calculate the similarity between the monitored pedestrian feature set T and the weighted multi-view fusion feature, b represents the dimension of SIFT feature, y represents the number of cameras arranged in the monitoring area, and one camera represents one view, n represents the number of target monitored pedestrians, and e represents the feature dimension;
[0035] Step E2: According to the pedestrian to be matched t mThe similarity between the weighted multi-view fusion features and the monitored pedestrian feature set T is as follows:
[0036]
[0037] Calculate the pedestrian t to be matched m The mismatch probability Q between the image at each angle and the target monitored pedestrian T m .
[0038] Furthermore, the aforementioned step G also includes the following steps G1-1 to G1-2;
[0039] Step G1-1: Calculate the matching similarity between the weighted multi-view fusion features of each primary pedestrian to be matched and the multi-view fusion features of each target monitored person, and obtain the maximum similarity between the weighted multi-view fusion features of each primary pedestrian to be matched under the main camera and the multi-view fusion features of the target monitored pedestrian. And the maximum similarity The corresponding target is the monitored pedestrian;
[0040] Step G1-2: According to the following formula:
[0041]
[0042] Determine the weighted multi-view fusion features of each primary pedestrian to be matched under the main camera and the pedestrian being monitored by the target d (1≤d≤n) Maximum similarity of multi-view fusion features Is it greater than or equal to the similarity threshold β of the pedestrian to be matched and the pedestrian to be monitored? th , if yes, the matching result between the pedestrian to be matched and the monitored pedestrian is that the pedestrian to be matched is the target monitored pedestrian; otherwise, the matching result is unsuccessful.
[0043] Furthermore, the aforementioned step G also includes the following steps G2-1 to G2-3;
[0044] Step G2-1: Calculate the matching similarity between the weighted multi-view fusion features of each primary pedestrian to be matched under the main camera and the multi-view fusion features of each target monitored person;
[0045] Step G2-2: Based on the matching similarity obtained in step G2-1, further obtain the matching similarity of the multi-view fusion features of each target monitored person and the weighted multi-view fusion features of each primary pedestrian to be matched under the main camera, sorted from large to small, and the maximum similarity between the multi-view fusion features of the monitored pedestrian and each primary pedestrian to be matched And the maximum similarity The corresponding target monitored pedestrian and the primary pedestrian to be matched;
[0046] Step G2-3: For each pedestrian to be matched and its corresponding maximum similarity The target monitored pedestrian is removed from the target monitored pedestrian with the maximum similarity. The target monitored pedestrian and the matching similarity between the weighted multi-view fusion features of each pedestrian to be matched under the main camera and the multi-view fusion features of each target monitored person are calculated, and it is judged whether the primary pedestrian to be matched and the target monitored pedestrian are completely eliminated. If yes, the maximum similarity is used The corresponding target monitored pedestrian and the primary pedestrian to be matched complete the matching of the pedestrian to be matched relative to the target monitored pedestrian, otherwise, return to step G2-1.
[0047] Taking advantage of the cameras arranged at multiple angles in the surveillance area, the main camera and the auxiliary camera are linked to intelligently extract weighted multi-view fusion features, reduce the impact of obstacles on the matching of specific target pedestrians, and reduce the probability of missed matching; the matching similarity is calculated based on the weighted multi-view fusion features and the multi-view fusion features of the monitored target pedestrian feature set, and then the mismatch probability is calculated. By setting the mismatch probability and similarity threshold conditions, it is hierarchically judged whether the pedestrian to be matched is the monitored pedestrian target, which reduces the probability of false matching. By calculating the matching similarity between the person to be matched and the target monitored pedestrian, and then finding the maximum similarity and eliminating the target monitored pedestrian, the person to be matched and the target monitored pedestrian can be accurately matched successfully. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 A flowchart of a specific pedestrian target matching method based on multi-view feature group intelligence fusion provided by the present invention;
[0049] Figure 2 The present invention provides a scene deployment schematic diagram of a specific pedestrian target matching method based on multi-view feature group intelligence fusion. DETAILED DESCRIPTION
[0050] In order to better understand the technical content of the present invention, specific embodiments are given and described as follows in conjunction with the accompanying drawings.
[0051] Various aspects of the invention are described herein with reference to the accompanying drawings, in which many illustrative embodiments are shown. The embodiments of the invention are not limited to those described in the accompanying drawings. It should be understood that the invention is implemented by any of the various concepts and embodiments described above, as well as the concepts and embodiments described in detail below, because the concepts and embodiments disclosed in the invention are not limited to any implementation. In addition, some aspects disclosed in the invention may be used alone or in any appropriate combination with other aspects disclosed in the invention.
[0052] like Figure 2As shown, based on the pedestrians in the target area and the multi-angle cameras covering various positions in the target area, the multi-angle camera selects k viewing angles for each pedestrian to shoot, and obtains the angle images of each pedestrian. According to the preset angle images corresponding to each target monitored pedestrian in the target area, each camera is powered by a solar panel, and each camera uploads the image to the cloud service through the 5G network and performs calculations. For each pedestrian to be matched in the target area, the following steps A to F are performed to complete the primary matching of the person to be matched, and then step G is performed to achieve the matching of the pedestrian to be matched relative to the target monitored pedestrian;,
[0053] Step A: For each target monitored pedestrian in the target area, based on each angle image of each pedestrian, obtain each angle image I of each target monitored pedestrian k , from each target monitored pedestrian image I k The SIFT features of images at each angle are extracted respectively {P 1k ,…,P nk}, and the spatial color histogram features of images at each angle {Z 1k ,…,Z nk},in, Where n represents the number of monitored pedestrians, i∈[1,n];
[0054] Then, the SIFT features of the monitored pedestrian angle images of each target are {P 1k ,…,P nk}, and the spatial color histogram features of images at each angle {Z 1k ,…,Z nk}, and then fuse to obtain the multi-view fusion features R of each monitored pedestrian i :
[0055]
[0056] Where b represents the dimension of SIFT features. According to the multi-view fusion features of each target monitored pedestrian, a target monitored pedestrian feature set T is constructed, where T = {R1,…,R n}, and enter it into the server, where n represents the number of target monitored pedestrians, and then execute step B.
[0057] Step B: For each angle image of the pedestrian to be matched, extract the pedestrian t to be matched from the camera y perspective m SIFT features of images at different angles in represents the feature value on dimension s, b represents the dimension of SIFT feature, y represents the number of cameras arranged in the monitoring area, and one camera represents one viewpoint; then the pedestrian t to be matched is extracted m Spatial color histogram feature information of images at different angles in, represents the number of pixels with absolute brightness value j; extract the viewpoint SIFT features and spatial color histogram feature information for each angle image, and treat the matched pedestrian t separately. m The multi-view SIFT features P of each angle image my and spatial color histogram feature Z my Fusion is performed to obtain the pedestrian t to be matched m The perspective fusion features corresponding to the images at each angle are: Wherein, b represents the dimension of SIFT features, y represents the number of cameras arranged in the monitoring area, and one camera represents one viewing angle. Then, step C is executed;
[0058] Step C: For each pedestrian to be matched t m Images of each angle, count each pedestrian t to be matched m The number of pixels N of each angle image y , 1≤y≤Y, Y represents the number of cameras deployed in the target domain; according to the following formula:
[0059]
[0060] Calculate each pedestrian t to be matched m The weight α of the angle image y ; and obtain the maximum weight α max ∈{α1,…,α Y} of the pedestrian t to be matched m , and further obtain the camera y corresponding to the pedestrian angle image to be matched with the largest weight. max ∈[1,Y], determine the camera's images of each angle for the pedestrian to be matched, count the number of pixels of the images of each angle for the pedestrian to be matched, and calculate the image weights of the images of each angle to obtain the angle image with the largest weight, and then further obtain the camera corresponding to the angle image; and determine the camera as the main camera in the target area; then execute step D;
[0061] Step D: Based on each pedestrian to be matched t m The angle image weight and the pedestrian t to be matched m Multi-view fusion features are fused according to the following formula:
[0062]
[0063] Obtain the weighted multi-view fusion features of the pedestrian to be matched under the main camera Wherein b represents the dimension of SIFT features, y represents the number of cameras arranged in the monitoring area, and one camera represents one viewing angle. Then, step E is executed;
[0064] Step E: Based on the pedestrian to be matched t m Weighted multi-view fusion features corresponding to the main camera And the multi-view fusion features R of each target monitored pedestrian in the monitored pedestrian feature set T i =(r i 1 ,…,r i b+256 ), i=1,…,n, according to the following formula:
[0065]
[0066] Calculate the similarity between the monitored pedestrian feature set T and the weighted multi-view fusion feature, b represents the dimension of SIFT feature, y represents the number of cameras arranged in the monitoring area, and one camera represents one view, n represents the number of target monitored pedestrians, and e represents the feature dimension; according to the pedestrian to be matched t m The similarity between the weighted multi-view fusion features and the monitored pedestrian feature set T is as follows:
[0067]
[0068] Calculate the pedestrian t to be matched m The mismatch probability Q between the image at each angle and the target monitored pedestrian T m . Then perform step F;
[0069] Step F: Determine the mismatch probability Q obtained in step E m Is it greater than or equal to the matching success threshold Q? th If yes, the initial matching result is unsuccessful, and the process returns to step B; otherwise, the initial matching result is successful, and the person to be matched constitutes the primary person to be matched; then the process goes to step G.
[0070] Step G: Calculate the similarity between each primary pedestrian to be matched and each target pedestrian to be monitored, and combine the preset similarity threshold β between the primary pedestrian to be matched and the target pedestrian to be monitored th , obtain the matching result between the primary pedestrian to be matched and the target monitored pedestrian.
[0071] According to the matching results of the primary pedestrian to be matched and the target pedestrian to be monitored, the present invention provides two subsequent matching processes: the first one: executing step G1-1 to step G1-2,
[0072] Step G1-1: Calculate the matching similarity between the weighted multi-view fusion features of each primary pedestrian to be matched and the multi-view fusion features of each target monitored person, and obtain the maximum similarity between the weighted multi-view fusion features of each primary pedestrian to be matched under the main camera and the multi-view fusion features of the target monitored pedestrian. And the maximum similarity The corresponding target is the monitored pedestrian;
[0073] Step G1-2: According to the following formula:
[0074]
[0075] Determine the weighted multi-view fusion features of each primary pedestrian to be matched under the main camera and the pedestrian being monitored by the target d (1≤d≤n) Maximum similarity of multi-view fusion features Is it greater than or equal to the similarity threshold β of the pedestrian to be matched and the pedestrian to be monitored? th , if yes, the matching result between the pedestrian to be matched and the monitored pedestrian is that the pedestrian to be matched is the target monitored pedestrian; otherwise, the matching result is unsuccessful.
[0076] The second type: Step G2-1 to Step G2-3;
[0077] Step G2-1: Calculate the matching similarity between the weighted multi-view fusion features of each primary pedestrian to be matched under the main camera and the multi-view fusion features of each target monitored person;
[0078] Step G2-2: Based on the matching similarity obtained in step G2-1, further obtain the matching similarity of the multi-view fusion features of each target monitored person and the weighted multi-view fusion features of each primary pedestrian to be matched under the main camera, and sort them from large to small. For example, there are 10 target monitored pedestrians and 5 primary pedestrians to be matched. Then, for each target monitored pedestrian, there are 25 similarities with the target pedestrian to be matched. Then sort these 25 similarities to obtain the maximum similarity between the multi-view fusion features of the monitored pedestrian and each primary pedestrian to be matched. And the maximum similarity The corresponding target monitored pedestrian and the primary pedestrian to be matched;
[0079] Step G2-3: For each pedestrian to be matched and its corresponding maximum similarity The target monitored pedestrian is removed from the target monitored pedestrian with the maximum similarity. The target monitored pedestrian and the matching similarity between the weighted multi-view fusion features of each pedestrian to be matched under the main camera and the multi-view fusion features of each target monitored person, that is, the person with the maximum similarity between the monitored target pedestrian and the primary pedestrian to be matched. And determine whether the primary pedestrian to be matched and the target monitored pedestrian are completely eliminated. If yes, the maximum similarity is used. The corresponding target monitored pedestrian and the primary pedestrian to be matched complete the matching of the pedestrian to be matched relative to the target monitored pedestrian, otherwise return to step G2-1, continue to perform similarity calculation, and sort again, and then find the next
[0080] Taking advantage of the cameras arranged at multiple angles in the surveillance area, the main camera and the auxiliary camera are linked to intelligently extract weighted multi-view fusion features, reduce the impact of obstacles on the matching of specific target pedestrians, and reduce the probability of missed matching; the matching similarity is calculated based on the weighted multi-view fusion features and the multi-view fusion features of the monitored target pedestrian feature set, and then the mismatch probability is calculated. By setting the mismatch probability and similarity threshold conditions, it is hierarchically judged whether the pedestrian to be matched is the monitored pedestrian target, which reduces the probability of false matching. By calculating the matching similarity between the person to be matched and the target monitored pedestrian, and then finding the maximum similarity and eliminating the target monitored pedestrian, the person to be matched and the target monitored pedestrian can be accurately matched successfully.
[0081] Although the present invention has been described above with preferred embodiments, it is not intended to limit the present invention. A person skilled in the art of the present invention may make various modifications and improvements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the definition of the claims.
Claims
1. A specific pedestrian target matching method based on multi-view feature group intelligence fusion, characterized in that: Based on the pedestrians in the target area and the multi-angle cameras covering various positions in the target area, according to the preset angle images corresponding to each target monitored pedestrian in the target area, the following steps A to F are performed for each pedestrian to be matched in the target area to complete the primary matching of the person to be matched, and then step G is performed to achieve the matching of the pedestrian to be matched with the target monitored pedestrian; Step A: for each target monitored pedestrian in the target area, extract the SIFT features and spatial color histogram feature information of each angle image corresponding to the target monitored pedestrian, and fuse them to obtain the multi-view fusion features of the target monitored pedestrian, that is, obtain the multi-view fusion features corresponding to each target monitored pedestrian, and construct the target monitored pedestrian feature set T; Then execute step B; Step B: for each angle image of the pedestrian to be matched, extract the view SIFT feature and the spatial color histogram feature information for each angle image, then fuse the view SIFT feature and the spatial color histogram feature information to obtain the view fusion features corresponding to each angle image of the pedestrian to be matched; Then execute step C; Step C: for each angle image of the pedestrian to be matched, the number of pixels of each angle image of the pedestrian to be matched is counted, and the image weight of each angle image is calculated to obtain the angle image with the largest weight, and then further obtain the camera corresponding to the angle image; and determine that the camera is the main camera in the target area; Then perform step D; Step D: Based on the image weights of each multi-angle image corresponding to the pedestrian to be matched and the multi-view fusion features of each multi-angle image, weighted processing is performed to obtain the weighted multi-view fusion features of the pedestrian to be matched under the main camera; Then perform step E; Step E: Based on the monitored pedestrian feature set T and the weighted multi-view fusion features of the pedestrian to be matched under the main camera, the similarity between the monitored pedestrian feature set T and the weighted multi-view fusion features is calculated; further, the mismatch probability between the images of the pedestrian to be matched at each angle and the monitored pedestrian feature set T is obtained; and then step F is executed; Step F: Determine the mismatch probability Q obtained in step E m Is it greater than or equal to the matching success threshold Q? th , if yes, the initial matching result is unsuccessful, and the process returns to step B; otherwise, the initial matching result is successful, and the person to be matched constitutes the primary person to be matched; Step G: Calculate the similarity between each primary pedestrian to be matched and each target pedestrian to be monitored, and combine the preset similarity threshold β between the primary pedestrian to be matched and the target pedestrian to be monitored th , obtain the matching result between the primary pedestrian to be matched and the target monitored pedestrian.
2. According to claim 1, a specific pedestrian target matching method based on multi-view feature group intelligence fusion is characterized in that: The multi-angle cameras covering various positions in the target area select k viewing angles to shoot each pedestrian, and obtain images of each pedestrian at various angles.
3. According to claim 2, a specific pedestrian target matching method based on multi-view feature group intelligence fusion is characterized in that: The step A comprises the following steps A1 to A3: Step A1: Based on the images of each pedestrian at each angle, obtain the images of each monitored pedestrian at each angle I k , and images of pedestrians under surveillance from various angles I k The SIFT features of images at each angle are extracted respectively {P 1k ,…,P nk }, and the spatial color histogram features of images at each angle {Z 1k ,…,Z nk },in, Where n represents the number of monitored pedestrians, i∈[1,n]; Step A2: Based on the SIFT features of each monitored pedestrian angle image, {P 1k ,…,P nk }, and the spatial color histogram features of images at each angle {Z 1k ,…,Z nk }, obtain the multi-view fusion features R of each monitored pedestrian i : Among them, b represents the dimension of SIFT features, Step A3: Based on the multi-view fusion features of each target monitored pedestrian, a target monitored pedestrian feature set T is constructed, where T = {R1,…,R n } and enter it into the server, where n represents the number of target monitored pedestrians.
4. According to claim 2, a specific pedestrian target matching method based on multi-view feature group intelligence fusion is characterized in that: The step B comprises the following steps B1 to B3: Step B1: Extract the pedestrian t to be matched from the camera y perspective m SIFT features of images at different angles in represents the feature value on dimension s, b represents the dimension of SIFT feature, y represents the number of cameras arranged in the monitoring area, and one camera represents one viewing angle; Step B2: Extract the pedestrian t to be matched m Spatial color histogram feature information of images at different angles in, represents the number of pixels whose absolute brightness is j; Step B3: Treat the matched pedestrians t separately m The multi-view SIFT features P of each angle image my and spatial color histogram feature Z my Fusion is performed to obtain the pedestrian t to be matched m The perspective fusion features corresponding to the images at each angle are: Among them, b represents the dimension of SIFT features, y represents the number of cameras deployed in the monitoring area, and one camera represents one viewing angle.
5. According to claim 2, a specific pedestrian target matching method based on multi-view feature group intelligence fusion is characterized in that: The step C comprises the following steps C1 to C2; Step C1: For each pedestrian to be matched t m Images of each angle, count each pedestrian t to be matched m The number of pixels N of each angle image y , 1≤y≤Y, Y represents the number of cameras deployed in the target domain; Step C2: According to the following formula: Calculate each pedestrian t to be matched m The weight α of the angle image y ;and Get the maximum weight α max ∈{α1,…,α Y } of the pedestrian t to be matched m , and further obtain the camera y corresponding to the pedestrian angle image with the largest weight. max ∈[1,Y], determine that the camera is the main camera.
6. According to claim 2, a specific pedestrian target matching method based on multi-view feature group intelligence fusion is characterized in that: The step D is specifically as follows: based on each image of the pedestrian to be matched t m The angle image weight and the pedestrian t to be matched m Multi-view fusion features are fused according to the following formula: Obtain the weighted multi-view fusion features of the pedestrian to be matched under the main camera Where b represents the dimension of SIFT features, y represents the number of cameras deployed in the monitoring area, and one camera represents one viewing angle.
7. According to claim 2, a specific pedestrian target matching method based on multi-view feature group intelligence fusion is characterized in that: The step E includes the following steps E1 to E2; Step E1: Based on the pedestrian to be matched t m Weighted multi-view fusion features corresponding to the main camera And the multi-view fusion features R of each target monitored pedestrian in the monitored pedestrian feature set T i =(r i 1 ,…,r i b+256 ), i=1,…,n, according to the following formula: Calculate the similarity between the monitored pedestrian feature set T and the weighted multi-view fusion feature, b represents the dimension of SIFT feature, y represents the number of cameras arranged in the monitoring area, and one camera represents one view, n represents the number of target monitored pedestrians, and e represents the feature dimension; Step E2: According to the pedestrian to be matched t m The similarity between the weighted multi-view fusion features and the monitored pedestrian feature set T is as follows: Calculate the pedestrian t to be matched m The mismatch probability Q between the image at each angle and the target monitored pedestrian T m .
8. According to claim 2, a specific pedestrian target matching method based on multi-view feature group intelligence fusion is characterized in that: The step G also includes the following steps G1-1 to G1-2; Step G1-1: Calculate the matching similarity between the weighted multi-view fusion features of each primary pedestrian to be matched and the multi-view fusion features of each target monitored person, and obtain the maximum similarity between the weighted multi-view fusion features of each primary pedestrian to be matched under the main camera and the multi-view fusion features of the target monitored pedestrian. And the maximum similarity The corresponding target is the monitored pedestrian; Step G1-2: According to the following formula: Determine the weighted multi-view fusion features of each primary pedestrian to be matched under the main camera and the pedestrian being monitored by the target d , where 1≤d≤n, the maximum similarity of multi-view fusion features Is it greater than or equal to the similarity threshold β of the pedestrian to be matched and the pedestrian to be monitored? th , if yes, the matching result between the pedestrian to be matched and the monitored pedestrian is that the pedestrian to be matched is the target monitored pedestrian; otherwise, the matching result is unsuccessful.
9. The specific pedestrian target matching method based on multi-view feature group intelligence fusion according to claim 2 is characterized in that: The step G also includes the following steps G2-1 to G2-3; Step G2-1: Calculate the matching similarity between the weighted multi-view fusion features of each primary pedestrian to be matched under the main camera and the multi-view fusion features of each target monitored person; Step G2-2: Based on the matching similarity obtained in step G2-1, further obtain the matching similarity of the multi-view fusion features of each target monitored person and the weighted multi-view fusion features of each primary pedestrian to be matched under the main camera, sorted from large to small, and the maximum similarity between the multi-view fusion features of the monitored pedestrian and each primary pedestrian to be matched And the maximum similarity The corresponding target monitored pedestrian and the primary pedestrian to be matched; Step G2-3: For each pedestrian to be matched and its corresponding maximum similarity The target monitored pedestrian is removed from the target monitored pedestrian with the maximum similarity. The target monitored pedestrian and the matching similarity between the weighted multi-view fusion features of each pedestrian to be matched under the main camera and the multi-view fusion features of each target monitored person are calculated, and it is judged whether the primary pedestrian to be matched and the target monitored pedestrian are completely eliminated. If yes, the maximum similarity is used The corresponding target monitored pedestrian and the primary pedestrian to be matched complete the matching of the pedestrian to be matched relative to the target monitored pedestrian, otherwise, return to step G2-1.
