Abstract: In this study, we propose an algorithm to detectro and find landmarks (corners) in 3-D point cloud. Point cloud classification is an approach to find road or specific target but it is usually a time-consuming task especially when theory of random field was introduced into this research area recent years. The proposed algorithm is adapted for fast preprocessing and its parameters don’t need online learning. The algorithm employs a fuzzy cluster method based on maximum entropy theory to segment points, then a multi-times weight least-square linear fitting algorithm is used to differentiate linear and nonlinear distributed point segment. We extract road surface instead of road boundary and filters are designed to find road area. A corner fitting method will find corners of buildings as land marks according to different distribution of points. The spatial dependences among different laser detectors are considered to refine the results of extracted features. Experiment results valid the algorithm. The algorithm successfully extracts road and corners of buildings in point cloud which is sampled from complex semi-structured environment.