My current work in stereo range data extraction is mostly involved with finding correspondence of points in the stereo images. The points are matched base on both spatial and temporal information.

	Left	Right
time 0
time 100

The input is a sequence of 200 images captured using BumbleBee stereo vision camera system. When I captured the images, I had a small light source I waved around the box object while it's being captured. Essentially the goal is to cause illumination variations on the surface over time. Then in my depth finding algorithm, I build a histogram of pixel intensity change over time and using this histogram as part of the comparison algorithm to find matching pixels between the stereo images. The results are noisy, but it’s an interesting start to explore in this direction. Some of the results with different search masks and sensitivities are shown below (intensity values represent relative depth):

3D Points Generated from Depth Image