An application of mutual information based stereo matching for ADAS

An important aspect of Advanced Driver Assistance Systems (ADAS) is the collision avoidance mechanism. The use of stereo matching techniques to attain this objective can prove to be very beneficial. The disparity maps obtained through stereo matching are a direct indication of the distance of the various objects present in the immediate vicinity. The data extracted from these disparity maps can be used for vehicular control. A data cost comprising of mutual information and a modified census transform is proposed for the generation of disparity maps. The proposed methodology involves converting the image section under consideration into a binary string which can be used for further processing. For this purpose, a new polyvalent census transform is proposed and its performance on stereo image data sets is discussed.