Abstract: For the reasons of the complex working environment and plenty of obstacles in facility agriculture, this paper studied the problem of identifying the obstacles by binocular vision recognition system.Binocular vision system was built up at first.OpenCV and Matlab were used for the binocular camera calibration respectively.Secondly, Bouguet algorithm was used to rectify the input images and analyse three kinds of matching algorithms.Then the smallest enclosing moment of the object could be extracted.At last, the distance between the objects was calculated and 3D reconstruction was used.The experimental results showed that the binocular vision system had high reliability, high real time, strong anti-interference ability and lower cost which can identify different obstacles with a wide application rage.It could satisfy the operational requirements of the unmanned control vehicle’s automatic obstacle avoidance in facility agriculture, and laid the foundation for vehicle obstacle avoidance and navigation.