To improve the performance of algorithm in real time and reduce the cost of system configuration, this paper presents a autonomous navigation method based on monocular vision for the application of mobile robot in semi-structured environment. The method includes image segmentation, information extraction and behavior decision-making. In order to solve the interference of the lights and shadows on the image segmentation, the method introduced similar measure, SIMILATION. By analyzing the boundary pixels of the results of segmentation, intersection information can be obtained. Similarly, by analyzing images from the robot closer, obstacle information can be acquired. After the image segmentation and information extraction, the method gives the behavior of decision-making based on the kinematic model of four-wheel independent drive mobile robot when it encounters obstacles or crossroads. At last, the simulation verified the effectiveness of the method.