ZHNet:Improved Deep Stereo Matching Network Based on PSMNet

Depth estimation of stereo images holds a crucial significance in the realm of computer vision. In most previous works, the result in week texture and edge regions are often unsatisfactory due to poor feature extraction. In this article, we propose a new network aimed at effectively improving the performance of feature extraction. Specifically, based on PSMNet model, we introduced the hysteresis attention mechanism and ASPP structure to optimize and improve the feature extraction process. We evaluated our algorithm with the KITTI2012 and KITTI2015 dataset. The results obtained from our experiment indicate that our method has superior performance compared to PSMNet on the KITTI2012 and KITTI2015 dataset.