Research On Visual Tracking And Motion Control Of Intelligent Vehicle Based On Embedded System Development

A complete intelligent car target recognition and tracking system based on image processing and embedded technology is designed to overcome the shortages in intelligent vehicle visual perception,such as low accuracy,poor real-time and weak anti-interference ability.An optimization algorithm which combined online template matching with shape detection is proposed to solve the problems like large error and low recognition rate in traditional template matching and single shape detection,and the algorithm also has a good performance in overcoming the tracking drift defect of dynamic template update.The route of vehicle's global path planning in tracking is optimized by the Q-reinforcement learning(Q-learning)algorithm,and radial basis function(RBF)network was brought in for generalization of action value functions to accelerate algorithm convergence speed.The main contents and the results of this paper are as follows:(1)Configuring the vehicle's hardware including camera,Bluetooth,and others based on the analysis of actual functional requirements.Designing the transplantation of embedded system to establish a complete intelligent vehicle experimental platform.(2)The video image is obtained through the V4L2 and it is transmitted in real time by using the TCP protocol Socket network model.Taking preprocessing on original image,such as graying,filtering and edge extracting based on digital image processing technology to obtain target image with high contrast and obvious contour edge.(3)An optimized target recognition algorithm based on template matching and shape detection is proposed.The algorithm optimize the template updating model and the matching traversal process in dynamic matching.Recognizing target information by target contour detection and calculating the target position coordinates through the centroid method and it improves the accuracy and real-time performance of target recognition and location.(4)On-line learning is carried out by Q-learning algorithm,and RBF network is brought in for generalization of action value functions to accelerate algorithm convergence speed and improve the learning efficiency in the path planning.The proportion integration differentiation(PID)algorithm is used to control the motor,and the actual tracking strategy is designed to make vehicle moving smartly and accurately.(5)The results show that the optimized target recognition and location algorithm can locate the target accurately,and the RBF network based on Q-learning algorithm effectively speeds up the convergence rate.the rotation angle curve of the two cars is basically coincided,and the change trend is always consistent.The embedded intelligent car system can not only track target accurately,but also has a stable performance in various aspects.Therefore,it overall reach an expected effect.