US 12,175,787 B2
Three-dimensional human pose estimation method and related apparatus
Weidong Chen, Shenzhen (CN); Wei Zhang, Shenzhen (CN); Di Kang, Shenzhen (CN); and Linchao Bao, Shenzhen (CN)
Assigned to TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED, Shenzhen (CN)
Filed by Tencent Technology (Shenzhen) Company Limited, Shenzhen (CN)
Filed on Sep. 1, 2022, as Appl. No. 17/901,700.
Application 17/901,700 is a continuation of application No. PCT/CN2021/098627, filed on Jun. 7, 2021.
Claims priority of application No. 202010734364.8 (CN), filed on Jul. 27, 2020.
Prior Publication US 2022/0415076 A1, Dec. 29, 2022
Int. Cl. G06V 40/10 (2022.01)
CPC G06V 40/103 (2022.01) 20 Claims
OG exemplary drawing
 
1. A three-dimensional human pose estimation method, performed by a computer device, the method comprising:
obtaining a video frame sequence of n views, the video frame sequence comprising a plurality of video frames sorted according to their timestamps, and n being an integer greater than 1;
extracting an initialization pose estimation result of each single video frame in the n views based on a neural network model;
performing single-frame and single-view human pose estimation on the initialization pose estimation result for each single video frame, to obtain n single-view pose estimation sequences respectively corresponding to the n views;
performing single-frame and multi-view human pose estimation according to single-view pose estimation results with the same timestamp in the n single-view pose estimation sequences, to obtain a multi-view pose estimation sequence; and
performing multi-frame and multi-view human pose estimation on each multi-view pose estimation result in the multi-view pose estimation sequence, to obtain a multi-view and multi-frame pose estimation result.