Research On Dynamic Obstacle Avoidance Of Binocular Vision Based On Artificial Potential Field Method

In recent years,with the rapid development of computer technology,the research of visual dynamic obstacle avoidance system of binocular is becoming more and more popular,and goes deep into China's political,economic,cultural,national defense,science and technology and education in all areas of fields.Based on this,a lot of enterprises according to their own internal split from the dynamic obstacle avoidance system for different levels of binocular vision applications,although to a certain extent,replace humans engage in production activities,but the algorithm itself exists some disadvantages.For example,the image feature matching process in the binocular vision system time-consuming and mismatches;in the process of dynamic obstacle avoidance,there are some problems,such as goals unreachable with obstacle nearby problem and the local minimum problem,which results in poor real-time ability and poor accuracy.In order to improve the efficiency of image matching of binocular vision system and increase the matching accuracy,solve the robot to be accurate in the process of moving obstacle avoidance problem to further improve the working efficiency of the robot,this paper made the following work:1.Based on the traditional binocular vision obstacle avoidance system,the Harris operator image matching algorithm and the SIFT image matching algorithm are analyzed and improved.SIFT operator to replace Harris operator to complete the image corner detection,improved the original SIFT image matching algorithm and the initial image feature point matching results after RANSAC(random sampling consistency algorithm)for image feature match was delayed,complete the image feature matching process.The matching efficiency and matching precision of the original image matching algorithm are improved,and the matching complexity of the image matching algorithm is reduced.2.The basic principle of the traditional artificial potential field method is studied,and the problem of the target point inaccessibility and the local minimum in the traditional artificial potential field method are analyzed.On the basis of the original algorithm,the repulsive potential energy field is redefined,and the influence scope of obstacles is segmented.Different repulsive potential energy functions are applied in different influence areas to solve the problem of unreachable target points in the original algorithm.Will improve after the repulsive force direction decomposition,in part,repulsive force direction tangent to the areas of obstacles,and with the gravity direction Angle is less than or equal to 90 °,another part of the repulsive force direction consistent with the direction of gravity,make its can not meet the conditions of the intelligent car trapped in local minimum,in order to solve the problems existing in the original algorithm.Experiment and data simulation of the improved algorithm are carried out to verify the effectiveness of the improved algorithm in the whole binocular vision obstacle avoidance system.The improved system can efficiently complete the stereo matching process for collecting images,and the whole system can finish the obstacle avoidance action in dynamic environment,and quickly restore to the original working state of the system.High matching efficiency,high accuracy and good real-time performance can meet the different needs of the dynamic obstacle avoidance system for binocular vision in different fields and industries.