200ms。 Obstacle detection is a key component of autonomous systems. A obstacle detection method combined monocular vision and stereo vision is studied for the vision navigated combine harvester. For monocular vision detection, H and S components are used to segment the image acquired by the left camera mounted on the combine harvester, and then through the fixed threshold value and binary processing the potential obstacle area is located. For stereo vision, the SIFT features are extracted from the potential obstacle area, and the ANN algorithm is utilized to get matching points. According to the obtained world coordinates the obstacle and the distance from the vehicle are calculated. In order to reduce the processing time the coefficient of image size linear transform is analyzed and it shows that the matching points are enough to satisfy the system need and the processing time is less than 200?ms when the coefficient is 4.0. The experiment using various mature wheat videos indicates that the method is valid to detect obstacles in front of the vehicle.