The basic algorithm will be to calibrate the cameras (do some pattern matching to determine the relative view points of each camera). Then perform stereomatching using a chessboard pattern or similar to determine the disparity map and convert that to the depth map by triangulation. I will upload some example photos which can be used to test.
Performance target is sub 1mm accuracy in both position and depth. And ability to process 1 depth map per second (2 source images). Resolution on the cameras is variable to meet both these targets. But the depth map must be able to achieve resolution of 1mm or less.
Code must be commented adequately enough for others to use.