The point cloud data should be clustered to ensure the accuracy and efficiency of 3D reconstruction work . According to the bottleneck of traditional clustering method when dealing with the mass data which produced by the stereoscopic vision system,this paper proposes an new method called density clustering algorithm based on bounding box. The algorithm first groups the given mass point cloud into lots of small clusters by using the bonding box,then completes the final clustering on the level of means of clusters.The result comes out that the new method is quite effective. It implements the simplification of the original point cloud,the simplified rate is as high as 96.75%, and finally divides the clusters into 5 classes.