Research On Obstacle Avoidance Technology Of Robot Vision Based On Optical Flow

Autonomous mobile robot is an important branch of robotics,which has been widely used in various fields of military,industry,environmental detection,emergency rescue,In order to achieve autonomous operation in a complex environment safely and smoothly,it's a basic function for the mobile robot to implement effective detection and avoidance of obstacles.The traditional way of using ultrasonic and infrared sensors to measure distance and avoid obstacles exists blind side.However the obstacle avoidance based on machine vision information could effectively avoid them,which has a higher reliability of obstacle avoidance.Therefore,a growing number of attention has been paid to the research on the theory and technology of robot visual obstacle avoidance in recent years,and the burgeoning method based on optical flow obstacle avoidance of machine vision has been the priority,because of its cognitive behavior closer to human beings and lower requirements on the acquisition equipment.However,due to its research has just started,there are still many technical bottlenecks needed to be further studied in practical applications.In view of above,the method based on optical flow obstacle avoidance will be emphasized on relevant research in this paper.Firstly,the LK(Lucas&Kanade)sparse optical flow algorithm is used to calculate the scene feature points' depth information namely collision time TTC(Time to contact),then clustering and analyzing for the distribution of feature points' TTC using DPFC(density peaks finding cluster).According to the results of TTC clustering,the obstacle region will be identified,and combined with the artificial potential field method,the obstacle will be avoided.The main research work is as follows:1.Research on LK optical flow algorithm and its improvement.In order to decrease the calculation amount in the process of obstacle avoidance,Shi-Tomasi method will be used to detect feature points in the visual scene.As to the obviously textural features of ground,for the sake of reducing the amount of calculation,toomany feature points will be eliminated according to the distribution of feature points on the ground.Then LK algorithm is used to calculate optical flow at the effective feature points,which proves the basis for the next step to calculate TTC.Concerning finite precision of the common gradient operator,a novel polynomial window function estimation algorithm was presented,for each pixel gray value with a polynomial function to represent the window,then Gaussian function would be used to compute the weight of each pixel to balance the influence of each point,and polynomial coefficients was acquired by using the least square method,then differentiating for the polynomial function to get the image gradient.This method can make better use of the global characteristics of the image,and the gradient has superior accuracy and robustness.The simulation results show that the improved algorithm presented in this paper could get more precise optical flow comparing to the common gradient operator,which lays the foundation for the accurate calculation of the TTC in the robot visual scene.2.Research on obstacle recognition based on optical flow clustering,a burgeoning and simple recently DPFC(density peaks finding cluster)algorithm was introduced in this paper which could implement two layers clustering for the optical flow information obtained by the LK algorithm: Firstly,the length of feature points' optical flow in the robot scene was clustered,which could judge and eliminate the error optical flow;Secondly,the TTC was calculated by using the optical flow at each characteristic point after the elimination of the error optical flow,then determine whether there were obstacles in the scene comparing to the pre-set security threshold TTC,if there were obstacles,then DPFC algorithm was used to cluster for the second level of TTC of each feature point to determine the cluster center,and the obstacle region and background would be distinguished according to the TTC value of the clustering center.The simulation results show that the DPFC algorithm introduced in this paper and implemented two layers clustering for optical flow information can accurately and efficiently identify the obstacle region in the visual scene.3.Research on obstacle avoidance strategy of optical flow based on artificialpotential field method.The artificial potential field method was introduced in this paper,and combined with obstacle recognition method based on optical flow and designed the obstacle avoidance strategy,the robot can complete the path planning.According to the clustering results of TTC above,the minimum kind of TTC value of each cluster center would be found and identified it on the potential obstacle that is nearest to the robot,then taken the peripheral feature points of this kind and the robot's connection as repulsive vector,the repulsive force field would be determined by calculating the repulsive force with the corresponding TTC of each point.Finally,the gravitational potential field will be calculated according to the position of target location,then the navigation and obstacle avoidance are realized effectively by utilizing the artificial potential field.4.An experimental platform for obstacle avoidance of optical flow based on monocular vision was built,an experimental study was carried out by selecting a specific scene in the campus,which verified the theoretical algorithm of the paper.The experimental results show that this robot obstacle avoidance method based on optical flow and artificial potential field method can effectively guide the robot to avoid obstacles and reach the destination successfully.