We have successful created a disparity map from our two Stereo images.  An initial estimate of the disparity between the left and right images was attained using a sum squared difference algorithm.  This algorithm attempts to match 5×5 blocks of pixels in the left image to the right image.  If a good enough match is not found a disparity of -1 is recorded.  After the initial disparity map is created we use the segmented image to preform a segmented based average on the disparities.  This serves to eliminate some of the outliers in the image and give us more consistent data.  The average is taken over a 11×11 pixel radius and pixels that are located outside the center pixels segment do not count heavily towards the average.  This allowing us to maintain differences between objects in the image while still averaging within them.  The whole process takes about 80msec in the GPU and the results of one run are shown below.  The picture of the disparity map shown below was colored to show depth, red is closest and blue is furthest away.

Left Rectified Image

Right Rectified Image

Segmented Image

Disparity Map