Segmentation and Tracking Multiple Objects under Occlusion from Multiview Video

Qian Zhang,King Ngi Ngan
DOI: https://doi.org/10.1109/tip.2011.2159228
IF: 10.6
2011-01-01
IEEE Transactions on Image Processing
Abstract:In this paper, we present a multiview approach to segment the foreground objects consisting of a group of people into individual human objects and track them across the video sequence. Depth and occlusion information recovered from multiple views of the scene is integrated into the object detection, segmentation, and tracking processes. Adaptive background penalty with occlusion reasoning is proposed to separate the foreground regions from the background in the initial frame. Multiple cues are employed to segment individual human objects from the group. To propagate the segmentation through video, each object region is independently tracked by motion compensation and uncertainty refinement, and the motion occlusion is tackled as layer transition. The experimental results implemented on both our sequences and other's sequence have demonstrated the algorithm's efficiency in terms of subjective performance. Objective comparison with a state-of-the-art algorithm validates the superior performance of our method quantitatively.
What problem does this paper attempt to address?