Stereo matching-based low-textured scene reconstruction for ALV navigation

Low-texture scene reconstruction using stereo matching for ALV navigation still remains challenging. In low-textured regions, local algorithms are guaranteed to fail and global algorithms are too time-consuming. Based on the Phong illumination model, a proposition that the low-textured region in an image is likely to correspond to a 3D plane is made in this paper. Then, local algorithms are used to matching the pixels in well-textured regions. On the contrary, a Bayesian model base matching method that combines a plane prior and pixel dissimilarity is designed for low-textured regions. Experimental results show that our algorithm gets good matching results in low-textured regions. It is effective both in urban environment and off-road enviroment with puddles, which proves the robustness of our algorithm. In addition, our algorithm can get real-time performance for a 320×240 image.