A multi-person motion capture method based on three-dimensional hypothesis space clustering

Abstract

The invention proposes a multi-person motion capture method based on three-dimensional hypothetical space clustering, which can be used for unmarked human body motion capture. The method includes: associating two-dimensional joint point candidate points between different views, reconstructing three-dimensional joint point candidate points, three-dimensional posture analysis and posture tracking. Without using a human body model or assuming any prior knowledge of the human body, the present invention can realize stable and credible two-dimensional and global three-dimensional human body posture estimation for multiple people of different shapes and variable numbers of people. The pose generated by the present invention satisfies multi-view geometric constraints and human bone length constraints, and realizes robust and credible human pose estimation in extremely challenging scenarios such as mutual occlusion and close interaction of multiple people.

Description

technical field [0001] The invention relates to a multi-person motion capture method based on three-dimensional hypothetical space clustering. Background technique [0002] According to different input data, the existing 3D human pose estimation methods can be divided into: based on monocular RGB images (sequences); based on depth images (sequences); and based on multi-view images (sequences). Estimating 3D human body pose based on monocular RGB images (sequences) is a problem with serious constraints. The observation input of the system is a complex natural image, and the state output is a high-dimensional human body pose. The process from observation input to state output is highly nonlinear. . Insufficient 3D pose training data sets, differences in the size and proportion of different human bodies, and the high dimensionality of the 3D pose space all make the reliability of 3D pose reconstruction a key problem to be solved. The 3D human pose estimation method based on d...

AI Technical Summary

Problems solved by technology

However, the 3D pose data set is difficult to obtain and the scale is small; at the same time, the state parameters of the 3D pose have a high dimensionality, and the individual body proportions and sizes vary greatly

This leads to the inability to guarantee the reliability of the 3D pose estimation results

Secondly, most 3D pose estimation methods only consider that there is only one person in the scene, and it is difficult to extend to multi-person scenes

Images