Stereoscopic video conversion based on depth tracking

Conventional 2D-to-3D conversion approaches always have some problems due to fast motion and non-rigid objects. In this proposed approach, a bi-direction optical flow is used to extract credible corner points from original sequences, and then Mean Shift algorithm helps to segment foreground object. In addition, improved active contour model offers more accurate outlines. Finally depth correlation gets to deal with the temporal depth variation caused by objects´ motion or camera motion. The experimental results give good performance with higher accuracy.