Stereo Matching Algorithm Research For Driver Assistance

Stereo matching is a very important component in computer vision and also plays a central role in stereo vision based Driver Assistance System. Driver assistance system is an early warning vehicle security system which can play a proactive role in driving. With two cameras installed in the vehicle to simulate the human visual system, system uses images obtained in the same time to run stereo matching to get disparity map. Then depth map can be obtained, which can help to recovery three-dimensional scene information. Opposed to radar, laser technology, stereo vision based driver assistance system provide drivers richer traffic information for initiative and early warning obstacle and identifying signs information.In Driver Assistance Systems, the vehicle is in high speed and the road environment is complex with much noise, so it is very important to find a matching method with high speed and good robustness to complex scenes and noise, etc. Image segmentation based stereo matching algorithm consider that smooth region in the original image always have smooth disparity and disparity discontinuities are generally appear with edge features. By using mean-shift image segmentation, this method get segmentation information and use them to fit initial disparity, which is got by fast window method, to get final disparity map. This method have good adaptability for complex information and road noise, this paper have in-depth research in this algorithm, mainly as follows: After in-depth analysis of image segmentation based stereo matching algorithm, in this paper, the initial disparity is computed by the combination of sum of absolute difference(SAD) algorithm and gradient (GRAD) algorithm. And the use of winner-win Strategy (with disparity obtained from the left and right respectively) can help to obtain more accurate initial disparity map. In the meantime, an adaptive threshold graph based image segmentation method is chosen to obtain the original information. The new segmentation-based stereo matching method not only get accurate depth map but also improve time efficiency. And experiment results show that the algorithm also has good robustness for the illumination or other noise in outdoor scenes, which can obtain the target depth accurately.Finally, the summary of our work is given and future direction for improvement is proposed.