Identification and navigation system of farmland path for high-clearance vehicle based on DM642

Abstract: This article presents a solution to achieve autonomous vehicle navigation and path recognition. The visual inspection system is established by utilizing agricultural four wheels drive vehicle equipped with a high performance digital media processor. This visual inspecting system is capable of collecting and displaying real time image, using an improved algorithm for path detection, which combines a serial of technologies, including 2G-R-B colour identification for identify the greens, central line algorithm for route calculation, and using double polyline algorithm including multi polyline to simulate Hough transform. In this article, the pro and con of each algorithm were compared and also its applicable environment by a series of path recognition and navigation tests. Although TMS320 DM642 has a strong operational ability, Hough transform needs a lot of operation. In order to reduce the delay in the process of image processing, a series of program optimizations had been launched in vehicle including Hough transform to extract path, Hough transform of double polylines fitting to extract path and Hough transform of multi polylines fitting to extract path. For example, the interesting areas were chosen to process, the step length of Hough transform appropriately increased, and interlaced scanning made the program of Hough transform in an easier, faster and more efficient way. The data receiving of the system was in the form of the serial port interrupt. In this agreement, according to the different environments, vehicle control chip could send proper instruction to control image processing chip to pick an optimal image algorithm. Meanwhile, according to the existing rules, the deviation of high-clearance vehicle's direction could be told to vehicle control chip by image processing chip in order to adjust the vehicle's direction appropriately according to the vehicle movement. Generally speaking, image processing chip would send data to vehicle control chip one time every 400 ms to control the vehicle in near real time. The accuracy could satisfy the need of the navigation. Experiment of walking straight showed that the standard deviation of lateral deviation was 7.199, the standard deviation of angular deviation was 6.294, and angular deviation could reach 22.5°. Experiment with velocity influence on navigation precision showed that the average tracking deviation was 0.61 cm and the maximum tracking deviation was 16 cm in small turn at low speed. The average tracking deviation was 5.21 cm and the maximum tracking deviation was 29 cm in small turn at high speed. The average tracking deviation was 0.78 cm and the maximum tracking deviation was 23 cm in large turn at low speed. The average tracking deviation was 6.36 cm and the maximum tracking deviation was 35 cm in large turn at high speed. Therefore, this navigation control system is stable and reliable when the vehicle passes a straight line. Navigation precision decreases under the condition of high speed. The effect of the tracing also decreases when the turning radius is too large. However, this system can meet the demands of agricultural vehicles in navigation. The results of a series of route recognition and navigation tests demonstrates the efficiency of this visual inspection system. The combination of different technologies such as central line algorithm, Hough transform, double polyline algorithm and multi polyline to simulate Hough transform can help the vehicle to achieve self-navigation in all kinds of indoor and outdoor environments. Also in a spiral of path recognition and navigation tests, the vehicle performs great adaptability and strong anti-interference ability, and also reacts quick and is highly stable. The application of such tracing system has full potential in the future.