Two-camera-based Human Motion Capture

Abstract

AbstractIn order to extract complex human motion precisely, multiple cameras are often used to capture the video sequences, then tracking and reconstruction of human motion can be achieved by virtue of the multi-view video sequences.[1] The self-occlusion problem that occurred during tracking can also be solved with the multi-view pattern. Multiple views mean that the same scene is captured with the same sampling rate from different viewpoints. According to the principle of vision, multiple corresponding image feature points are competent for reconstructing 3D coordinates of feature points accurately. Therefore, compared with monocular video sequence, 3D reconstruction is easier under multi-viewpoints. However, difficulties of feature correspondence and self-occlusion also exist in the tracking of multi-view video sequences. Especially, the automatic corresponding of multiple views is still a challenging issue. In this chapter, we intend to give readers more insight into the two-camera-based human motion capture as well as the VBHAS V3.0 (Video-based Human Animation).KeywordsKalman FilterFeature PointTracking ResultGravity CenterAdjacent FrameThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.