Abstract: Automatic navigation technology of agricultural machinery is concerned in the field of intelligent agriculture, while the obstacle detection is an important part. The shortcomings of sensor detection technology in the early years were analyzed, and the application and application prospects of computer vision in agricultural machinery intelligent obstacle detection methods were summarized. The results show that sensor technology is now developed from single sensor to multi-sensor fusion at very mature stage, but it is still affected by obstacles surface and too high detection cost. In recent years, computer vision and deep learning have been used in agriculture of convolutional neural network, but it should be improved much in occlusion, remote detection, moving obstacle detection and other aspects. According to the agricultural obstacle detection techniques in the past 20 years, the existing problems are summarized, and a new idea for the future development of smart agriculture in China is proposed.