Vision-based Road Detection And Navigation For AGV Systems

AGV system plays an important role in manufacture automatization:By applying the AGV system, manufacturers can effectively reduce labor costs, and improve their production. In recent years, the development of more powerful computers and sensors brings the rise of research in vision-based navigation algorithms, which makes more and more intelligent vehicles come into reality. This paper makes research on road detection and following methods of AGV system in outdoor environment, to implement a reliable, efficient navigation system for outdoor AGVs.The research consists of four parts:the designing and implementing of vision-based road detection, the designing and implementing of vision-based road following, the designing of AGV’s hardware system, and the designing and implementing of AGV’s software system.Based on research on road detection methods, a self-adaptive road detection method based on vision and cluster analysis is designed and implemented in this paper (chapter2). By applying the automatic selection of samples, the vehicle can add samples dynamically without manual operations. The proposed method has no drawbacks as in traditional road detection methods based on learning algorithm:the selection of samples needs large amount of manual operations. Verified by experiments on our self-designed AGV system, the method can effectively reduce the impact from light, shadow, lane marks to the road detection, and it is adaptive to the changes of road conditions, and need much less manual operations.A road following method is proposed in this paper:based on the result of road detection, the method extracts the two boundaries of road area, and uses high order curves to fit these boundaries, after that,the central line of the road is estimated based on which the moving strategy is made. Kalman filter is applied to the estimation of central line to make the line change smoothly in consecutive frames.In this paper, we also design and implement a novel AGV hardware system, which uses multiple kinds of sensors. Compared to AGVs with single kind of sensors, our system is more reliable. Based on the hardware system, we also develop a software system for running and verifying our algorithm.At last, this paper analyzes the drawbacks of the proposed methods, and shows how to improve the methods in subsequent research.