Stereo matching is the key and challenging problem of binocular stereo vision and it restricts the applica-tion of binocular stereo vision． Noises，obstacles and textureless region reduce the matching accuracy． In this pa-per，a laser speckle based stereo matching algorithm was proposed to solve these problems． Based on the gradient orientation of each pixel，a new matching cost function was proposed． Then semi-global matching and guided filter were cascaded to aggregate matching cost． This method can meet the real-time requirements and improve the adap-tive capacity of algorithm for slant plane． Finally，a confidence detection method was proposed to refine the disparity in obstacle regions and the edges regions based on the mean value and minimum value of the aggregated matching cost． Its application in logistics industry shows that the proposed algorithm can be used to measure the size of pac-kage and the accuracy can reach to ±5%．