Robust Stereo Data Cost With a Learning Strategy

摘要: The performance of stereo matching algorithms strongly depends on the quality data/matching cost. Most state-of-the-art data costs require expert knowledge for design a transformation function, such as census handling gray-level changes monotonically, adaptive normalized cross correlation Lambertian cases, guided filtering preserving edge information, and local density encoding illumination differences. However, it is difficult to complex function handle unknown factors that often occur in driving conditions snow, rain, sun. Therefore, this paper has investigated deep learning strategy develop novel cost model without using much knowledge. Experimental results show proposed obtains better than judged by standard KITTI benchmark, Middlebury, HCI datasets.