US 12,236,685 B2
Pedestrian search method, server, and storage medium
Yunyun Pan, Shenzhen (CN); Dongfang Li, Shenzhen (CN); and Xiaofei Mao, Shenzhen (CN)
Assigned to ZTE CORPORATION, Shenzhen (CN)
Appl. No. 17/784,989
Filed by ZTE CORPORATION, Shenzhen (CN)
PCT Filed Dec. 16, 2020, PCT No. PCT/CN2020/136905
§ 371(c)(1), (2) Date Jun. 13, 2022,
PCT Pub. No. WO2021/129491, PCT Pub. Date Jul. 1, 2021.
Claims priority of application No. 201911354606.4 (CN), filed on Dec. 25, 2019.
Prior Publication US 2023/0012137 A1, Jan. 12, 2023
Int. Cl. G06K 9/00 (2022.01); G06T 7/215 (2017.01); G06V 10/74 (2022.01); G06V 20/40 (2022.01); G06V 20/52 (2022.01); G06V 40/20 (2022.01)
CPC G06V 20/53 (2022.01) [G06T 7/215 (2017.01); G06V 10/74 (2022.01); G06V 20/46 (2022.01); G06V 20/52 (2022.01); G06V 40/25 (2022.01); G06V 2201/07 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A pedestrian search method, comprising:
performing a pedestrian detection on each segment of monitoring video to obtain a plurality of pedestrian tracks, wherein each pedestrian track of the plurality of pedestrian tracks comprises a plurality of video frame images of a same pedestrian; and
determining pedestrian tracks belonging to the same pedestrian according to video frame images in the plurality of pedestrian tracks, and merging the pedestrian tracks of the same pedestrian;
wherein the determining the pedestrian tracks belonging to the same pedestrian according to the video frame images in the plurality of pedestrian tracks comprises:
extracting N video frame images from the each pedestrian track;
combining the plurality of pedestrian tracks in pairs, wherein two pedestrian tracks in each group of pedestrian tracks are respectively recorded as a first pedestrian track and a second pedestrian track;
combining N video frame images extracted from the first pedestrian track and N video frame images extracted from the second pedestrian track in pairs to obtain a plurality of image combinations;
matching two video frame images in each image combination of the plurality of image combinations, and obtaining a matching result corresponding to each image combination of the plurality of image combinations; and
determining whether the two pedestrian tracks belong to the same pedestrian or not according to a plurality of matching results, and
wherein the plurality of matching results comprise a plurality of similarities; and
the matching the two video frame images in the each image combination of the plurality of image combinations, and the obtaining the matching result corresponding to the each image combination of the plurality of image combinations comprises:
extracting feature data from the two video frame images in the each image combination of the plurality of image combinations by utilizing a pedestrian re-identification network with a shielding analysis, respectively; and
obtaining a similarity corresponding to each image combination of the plurality of image combinations according to the feature data of each video frame image of the two video frame images.