Abstract: To deal with the confusion of foreground and background disparity at the edge, an edge-aware stereo matching method is proposed. Cascaded census transform is used to enhance the anti-noise ability, during the cost compute stage. SLIC Super-pixel segmentation information is adopted to fast edge-preserving cost aggregation at the cost aggregation stage. The edge disparity is further optimized by the guide cross filter at the stage of disparity post-processing. Experimental results show that the method obtain high quality disparity results in the Middlebury test sets and the actual scene, and the disparity at the edge is improved compared with the previous non-local stereo matching method. The experimental also found that guided cross filter proposed in this paper can solve the transition problem of the point cloud at the edge in the point cloud upsampling technique.