Research On Key Technology Of Moving Object Detecting And Tracking Based On Mobile Robot In Dynamic Environment

Moving target detecting and tracking is a hot issue in the field of robotics research.Accurate and reliable target tracking technology can be widely used in military,transportation, and scientific research. However, due to the complexity of environment,such as lighting, occlusion and other factors, as well as the diversity of the target itself,target detection and tracking technology has met great difficulties. In dynamicenvironment, since the state of the surrounding environment is not clear, the function ofobstacle avoidance should be implemented at the same time as the function of trackingmoving target.Based on the anylasis of traditional visual target tracking algorithm, this articleproposes a new mobile target tracking method combined with Kalman filteringtechnique. Also, we add obstacle avoidance function to robot in order to reduceenvironmental restrictions and expand the range of tracking. The main work of thispaper can be summarized as follows.In order to solve the robot obstacle avoidance problem in the dynamic environment,we design an obstacle avoidance scheme based on BP neural network. Firstly, itimproves BP neural network training mode, solving the problem of long training cycleand low efficiency problem effectively. Further, to meet the real-time demand ofobstacle avoidance, this article designs a hardware accelerator for obstacle avoidancealgorithm based on FPGA. Experiments show that, the improved algorithm can reducetraining time of neural network, and system response becomes faster. Compared withsoftware version running on a desktop PC, the hardware accelerator designed in thispaper can reach up to100X speedup.To solve the problem of moving target detecting and tracking, we use the L-Koptical flow algorithm to establish the optical flow field firstly.Then we applied theoptical flow field to track the moving target to shorten the time of the search targeteffectively, by estimating velocity and trajectory of the target based on the optical flowfield. Taking into account the need of calculating optical flow field for the whole image,we accelerated the L-K optical flow algorithm based on FPGA to meet the real-timedemand.There are several issues in traditional CamShift algorithm, such as target missingdue to color interference and occlusion. To solve these problems, we propose anautomatic tracking algorithm combined CamShift algorithm with Kalman filter. Theequations of robot motion and optical flow field are treated as the input of the Kalmanfilter, and the Kalman filter can effectively predict the target trajectory, and furtherreduce the search area of the target. As a result, the computational complexity isreduced and tracking efficiency is improved. Experiment results indicate that the improved algorithm we proposed can fully meet the real-time requirement of movingtarget tracking.