Disparity prediction between adjacent frames for dynamic scenes

Real-time stereo matching in image sequences is important in video monitoring, robotic navigation and intelligent vehicle, etc. Spatiotemporal stereo and scene flow can be used to produce temporally coherent disparity of dynamic scenes. However, most methods do not use the previous disparity map sufficiently to compute the current one. Thus, the disparity range limits the speed of disparity computation for each stereo pair. This paper integrates the temporal information into the stereo computation, and presents the relationship between consecutive disparity maps, which makes the disparity prediction reasonable. The scheme can produce a sequence of temporally coherent disparity maps rapidly. The tests performed on simulated and real stereo sequences confirm the validity of our approach.