CN-121982653-A - Passenger flow statistics method and device, computer equipment and readable storage medium

CN121982653ACN 121982653 ACN121982653 ACN 121982653ACN-121982653-A

Abstract

The application provides a passenger flow statistical method, a passenger flow statistical device, computer equipment and a readable storage medium, wherein personnel track data containing the same target multi-frame detection result are obtained, the multi-frame detection result is subjected to preliminary screening based on a space-time relation between a preset geometric condition and an adjacent detection result to obtain a preliminary screening detection result, quality scores are calculated and sequenced for the preliminary screening detection result, a preset number of target detection results are selected according to the sequencing result, static humanoid areas are formed by accumulating continuous static detection results in different tracks, the target detection results are filtered according to the static humanoid areas to obtain a to-be-identified personnel image, feature vectors of the to-be-identified personnel image are extracted, re-identification matching is carried out on the feature vectors in a corresponding feature matching space according to the behavior type of the personnel track, and the passenger flow statistical result is generated based on the re-identification matching result and personnel attribute information. By adopting the method, the accuracy and stability of passenger flow statistics can be greatly improved.

Inventors

HE JIE
SONG XIAOWAN
LI JINLIAN
DING ZHAOFENG
ZHOU XIANGGEN
GU XIANGYU
YU JIALIN
WANG HENG

Assignees

杭州杰峰软件有限公司

Dates

Publication Date: 20260505
Application Date: 20260403

Claims (10)

1. A method of passenger flow statistics, the method comprising: acquiring personnel track data containing multi-frame detection results of the same target; Performing preliminary screening on the multi-frame detection results based on a space-time relationship between preset geometric conditions and adjacent detection results to obtain preliminary screening detection results; calculating quality scores of the primary screening detection results, sorting the quality scores, and selecting a preset number of target detection results according to the sorting results; Accumulating to form a static humanoid region based on the detection results of continuous stillness in different tracks, and filtering the target detection results according to the static humanoid region to obtain a personnel image to be identified; Extracting the feature vector of the personnel image to be identified, and re-identifying and matching the feature vector in a corresponding feature matching space according to the behavior type of the personnel track; and generating a passenger flow statistical result based on the re-identification matching result and the personnel attribute information.
2. The method according to claim 1, wherein the preliminary screening of the multi-frame detection results based on the space-time relationship between the preset geometric condition and the adjacent detection results to obtain a preliminary screening detection result comprises: removing detection results of size abnormality according to a preset detection frame length-width ratio threshold; calculating the cross ratio between adjacent frame detection frames; Comparing the cross ratio with a preset threshold value, and judging whether the adjacent detection frames are relatively static; Filtering the test results determined to be relatively stationary to obtain the primary screening test results.
3. The method of claim 1, wherein calculating a quality score for the primary screening test results and ranking the quality scores, and selecting a predetermined number of target test results based on the ranking results, comprises: Calculating the quality score of the detection frame according to a preset formula according to the height and the width of the detection frame in each primary screening detection result; Ranking all of the primary screening test results in descending order based on the quality scores; and selecting the highest predetermined number of results from the arranged results as the target detection results.
4. The method of claim 1, wherein the accumulating the static humanoid region based on the continuously stationary detection results in the different trajectories includes: Identifying a detection result with the space kept unchanged in a plurality of continuous frames, and recording the center point of the detection result as a static point candidate; Accumulating the static point candidates from different tracks according to the space coordinates, and merging the static points adjacent to each other in space; and when the duration of the combined static points exceeds a preset threshold value, generating a rectangular area by taking the combined static points as the center, and taking the rectangular area as the static humanoid area.
5. The method according to claim 1, wherein the filtering the target detection result according to the static humanoid area to obtain the image of the person to be identified includes: Judging whether the center point of each target detection result is positioned in the static humanoid area; if the center point is positioned in the area, rejecting the target detection result; and cutting the image of the person corresponding to the reserved target detection result to be used as the image of the person to be identified.
6. The method according to claim 1, wherein the re-recognition matching of the feature vector in the corresponding feature matching space according to the behavior type of the person trajectory includes: Determining the behavior type of the person according to the person track; calculating the similarity between the feature vector and the historical feature vector in a feature matching space corresponding to the behavior type; And carrying out identity judgment based on the similarity, and outputting a judgment result as a matching result.
7. The method of claim 1, wherein generating the passenger flow statistics based on the re-identification matching results and the personnel attribute information comprises: Distributing or associating the identity mark for the personnel based on the re-identification matching result; Acquiring personnel attribute information through an attribute identification model; and combining the identity mark and the personnel attribute information to generate a passenger flow statistical result.
8. A passenger flow statistics apparatus, the apparatus comprising: The personnel track data acquisition module is used for acquiring personnel track data containing the multi-frame detection result of the same target; The primary screening detection result determining module is used for carrying out primary screening on the multi-frame detection results based on a space-time relationship between a preset geometric condition and adjacent detection results to obtain primary screening detection results; the target detection result determining module is used for calculating quality scores for the primary screening detection results and sorting the quality scores, and selecting a preset number of target detection results according to the sorting results; The to-be-identified personnel image determining module is used for accumulating to form a static humanoid region based on the detection results of continuous stillness in different tracks, and filtering the target detection results according to the static humanoid region to obtain to-be-identified personnel images; the re-recognition matching module is used for extracting the feature vector of the personnel image to be recognized, and re-recognition matching is carried out on the feature vector in the corresponding feature matching space according to the behavior type of the personnel track; and the passenger flow statistical result generation module is used for generating passenger flow statistical results based on the re-identification matching result and the personnel attribute information.
9. A computer device comprising a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, the processor and the memory in communication via the bus when the computer device is in operation, the machine-readable instructions when executed by the processor performing the steps of the passenger flow statistics method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a computer program which, when run by a processor, performs the steps of the passenger flow statistics method according to any of claims 1 to 7.

Description

Passenger flow statistics method and device, computer equipment and readable storage medium Technical Field The present application relates to the field of computer vision, and in particular, to a passenger flow statistics method, apparatus, computer device, and readable storage medium. Background With the rapid development of intelligent video analysis technology, passenger flow statistical methods based on personnel detection and re-identification have become core technical support for scenes such as store operation optimization, market flow regulation and control, public place management and the like. The technology does not need manual intervention, can automatically complete capturing of personnel targets, feature extraction and identity association through video acquisition equipment, efficiently output key data such as passenger flow quantity, in-out behaviors and the like, provides a data-driven decision basis for scene managers, and is continuously expanding the application range to various personnel-intensive places. The existing mainstream passenger flow statistics technology takes a single frame or a simple time sequence segment as a core processing unit, and the core thought is to extract the visual characteristics of people at different time points and perform global similarity comparison, so that identity records of the same person are associated, and passenger flow counting and behavior analysis are finally realized. However, researches find that in the prior art, since the multi-frame detection results in the continuous track of the same person are not subjected to systematic quality control, a large number of low-quality detection results including shielding, body shooting, morphological distortion or high repetition directly participate in the feature extraction and re-identification processes, so that the calculation load and storage cost of the system are greatly increased, invalid noise is introduced into the feature matching process, the stability of personnel identity association is reduced, the deviation of passenger flow statistics results is finally caused, and the core requirement of scenes on statistical accuracy is difficult to meet. Disclosure of Invention Accordingly, the present application is directed to a passenger flow statistics method, apparatus, computer device and readable storage medium, which can greatly improve the accuracy and stability of passenger flow statistics. In a first aspect, an embodiment of the present application provides a passenger flow statistics method, where the method includes: acquiring personnel track data containing multi-frame detection results of the same target; Performing preliminary screening on the multi-frame detection results based on a space-time relationship between preset geometric conditions and adjacent detection results to obtain preliminary screening detection results; calculating quality scores of the primary screening detection results, sorting the quality scores, and selecting a preset number of target detection results according to the sorting results; Accumulating to form a static humanoid region based on the detection results of continuous stillness in different tracks, and filtering the target detection results according to the static humanoid region to obtain a personnel image to be identified; Extracting the feature vector of the personnel image to be identified, and re-identifying and matching the feature vector in a corresponding feature matching space according to the behavior type of the personnel track; and generating a passenger flow statistical result based on the re-identification matching result and the personnel attribute information. Optionally, the preliminary screening of the multi-frame detection result based on the space-time relationship between the preset geometric condition and the adjacent detection result to obtain a preliminary screening detection result includes: removing detection results of size abnormality according to a preset detection frame length-width ratio threshold; calculating the cross ratio between adjacent frame detection frames; Comparing the cross ratio with a preset threshold value, and judging whether the adjacent detection frames are relatively static; Filtering the test results determined to be relatively stationary to obtain the primary screening test results. Optionally, the calculating the quality score for the primary screening test results and sorting the primary screening test results, and selecting a predetermined number of target test results according to the sorting results, including: Calculating the quality score of the detection frame according to a preset formula according to the height and the width of the detection frame in each primary screening detection result; Ranking all of the primary screening test results in descending order based on the quality scores; and selecting the highest predetermined number of results from the arranged results as the target detect