Abstract: Accurately extracting the point cloud collection of the target to be measured in the three-dimensional point cloud data is a key issue of the three-dimensional point cloud target recognition technology, and it is also an important challenge in the field of target recognition from 2D to 3D in recent years. The main difficulty is to quickly find the correlation function relationship between discrete point clouds. Combining stereo vision and feature matching, a constraint mechanism of target point cloud that can characterize different field of view conditions is constructed, and the original feature matching algorithm is optimized by using stereo vision as the constraint condition. An estimation algorithm based on stereo vision is designed, and recognition rules under different selection ratios are obtained through training and learning. The experiment uses ARIES Lidar to collect point clouds, and selects three typical target states through MATLAB. When the target discrimination is high, the target recognition rate before and after optimization is above 98%. When the target discrimination is low, the restriction conditions of the target boundary after optimization can improve the recognition probability. The position deviation of the optimized point cloud data can reach 0.55 mm, which is 0.19 mm higher than 0.74 mm before optimization. At the same time, the convergence time curve of the optimized algorithm is better than before optimization. The average convergence time above 3000 points is about 8.33 s, which is better than 12.76 s before optimization. In summary, the optimized algorithm has better recognition efficiency.