In this paper, we study the technology of the robot environment map building based on laser radar scanner (SLAM). Aiming at the problem of high computation and high dimension matching in SLAM technology, this paper takes the method of reducing resource consumption and improving data association as the research starting point, and puts forward the framework of SLAM algorithm. Because of the high density environmental characteristics will lead to map growth with high rate. In the first part, this paper extracts segment descriptors from the point cloud data of laser radar scanners to establish of local sparse feature map. Then, extract the Matching feature. The Matching feature will be used for registration and prediction of the robot's view measurement and use the registration results to correct dead reckoning robot model, getting more accurate robot pose. Finally, the position and the observation prediction are used to achieve the Extend Calman filter to obtain the final position and attitude estimation.