Abstract: In order to realize obstacle recognition of unmanned ground vehicle(UGV) in unpredicted environment, and to improve the accuracy of recognition, a system based on three-dimensional radar is proposed. A classification algorithm is given according to its function, angle and distance are used as two thresholds to classify data. Then the blocks which impede the vehicles can be detected. The reliability of the system and the veracity of the algorithm are verified through experimentation. The system can provide a real-time evaluation method in unpredicted environment for intelligent vehicle.