Multi-view Human Motion Capture Method Based on Human Motion Prediction

Abstract

The invention discloses a multi-view human motion capture method based on human motion prediction. The method performs three-dimensional human body reconstruction on pictures synchronously taken by multiple cameras with different angles of view to obtain a three-dimensional skeleton of each human body; for subsequent frames, according to the previous frame The reconstructed 3D skeleton gives prediction results and confidence for the 3D key point position of the human body in the current frame; runs the human body detector for the key frame pictures to detect the 2D bounding box of each human body; for non-key frame pictures , projecting the three-dimensional skeleton of the motion prediction of the previous frame into the image of each view angle of the current frame, quickly obtaining the two-dimensional bounding box of the human body under each view angle of the current frame, thereby reducing the overhead of the human body detector and improving the algorithm efficiency; the present invention simultaneously The time series information is used to calculate the visibility of the key points of the human body, and the accuracy of the 3D skeleton reconstruction of the human body is improved based on the visibility.

Description

technical field [0001] The invention belongs to the technical field of computers, and in particular relates to a multi-view human motion capture method based on human motion prediction. Background technique [0002] Multi-view human motion capture refers to the restoration of the 3D motion of the human body based on multi-view videos. Most of the current related methods include two stages of detection and reconstruction. Although good results have been achieved on public datasets, their detection and reconstruction modules is separated, the detector cannot benefit from the result of previous frame reconstruction, and lacks motion prediction. Also, the visibility of keypoints is not well exploited. Contents of the invention [0003] The purpose of the present invention is to address the deficiencies in the prior art, and propose a human body detection method based on motion prediction, and use timing information to calculate the visibility of each key point, so as to impro...

AI Technical Summary

Problems solved by technology

[0002] Multi-view human motion capture refers to the restoration of the 3D motion of the human body based on multi-view videos. Most of the current related methods include two stages of detection and reconstruction. Although good results have been achieved on public datasets, their detection and reconstruction modules are separated, the detector cannot benefit from the results of previous frame reconstruction, and lacks motion prediction

Also, the visibility of keypoints is not well utilized

Images