In order to identify foliage and the adjacent obstacles in the vegetation scenes, an object detection algorithm of three-dimensional laser radar is proposed. The neighborhood characteristics of neighboring points are constructed in point cloud, and new characteristic parameters are extracted as determining criterion. Then the Gaussian mixture model is obtained by using the maximum expectation algorithm to characterize the distribution of the parameters. Finally, the priori model is established by using Markov random field. The optimal objective function is obtained by the graph-cut method under the maximum posteriori probability framework. This algorithm has been successfully applied to the unmanned platform. The experimental results show that the algorithm can effectively identify foliage and their adjacent obstacles, and the boundaries of obstacles can be detected clearly. Compared with traditional algorithms, the proposed algorithm is more robust and accurate, and its response time meets the demand of practical applications.