Deep Stereo Network with Cross-Correlation Volume Construction and Least Square Aggregation

Abstract Stereo matching is of great importance in robot operation, autonomous driving and virtual reality. Large textureless and depth discontinuity regions are still the problems that limit the accuracy of stereo matching tasks. Traditional correlation-based volumes only measure the feature similarity within the same channel of the feature maps, which lead to insufficient feature similarity learning between different channels and perform poorly in large textureless regions with high feature similarity requirements. To address the problems in large textureless regions, we propose the cross-correlation based cost volume construction which adequately learn the feature similarity in different channels of the feature maps. To address the problems in depth discontinuity regions and other gradient sensitive regions, we propose the differentiable least square aggregation module which can sufficiently utilize the gradient information and enhance the aggregation ability of the cost aggregation network for gradient features. Extensive experiments show that the proposed method solve the problems effectively in the above difficult regions and achieve the state-of-the art performance on Scene Flow dataset, KITTI 2012 dataset and KITTI 2015 dataset.