Abstract: In order to detect the obstacles from the disparity image correctly and resist the interference of the noise robustly,a fast and robust obstacle detection algorithm for off-road autonomous mobile robots is proposed.Firstly,a complete definition of obstacle point integrating the restrictions of height,continuity and slope is given,and then the idea of region growing is used to detect the obstacles with these restrictions.Finally,the algorithm is optimized by using the redundancy detection method and introducing the multi-dimensional strategy.In addition,the principle of the redundancy detection method is proved.Experiment results show that the algorithm is effective and robust in different environments.The average runtime of the algorithm is 25ms,which is fast enough for real time application.