Aiming at the limitations of single sensor in environment perception for intelligent vehicles, a method of detecting obstacles based on information fusion from camera and laser radar was proposed for autonomous agricultural vehicles. For the images captured from monocular camera, significance detection was carried out by using Ft algorithm and the significance images were generated. Meanwhile, cluster analysis based on data correlation assessment was conducted for reflection data points from laser radar to determine the priori information such as the number, boundary and location of obstacles. Then the pixel points corresponding to the laser radar data points were regarded as the seed points, and the significance images generated were activated by the seed points. Lastly, the region segmentation based on the region growth method was implemented to obstacles. The experimental results showed that the image significance detection based on Ft algorithm had a better edge detection effect, and the region growth method based on the seed points can effectively segment the obstacles. The information fusion of machine vision and laser radar can better eliminate the interference of non-obstacles and achieve the complete detection of obstacles.