Hardware friendly algorithm of HR real time stereo matching for automatic drive

Stereo matching is a most challenging topic in artificial intelligence. Local algorithms are better than global ones with lower complexity and higher speed for HD real-time stereo matching. This paper proposes an optimized local stereo algorithm adaptive window disparity estimation with cross check (AWDE-CC). A region growth algorithm to fill bad pixels is also proposed which can bring an improvement of the performance in low texture regions efficiently. The average error rate of the generated disparity maps is approximately 8%.