Obstacle Avoidance Path Planning Of Autonomous Navigation Agricultural Machinery Based On Machine Vision

With the rapid progress of science and technology,leading technology is more and more applied to modern agricultural production.Under the condition of no man-made control,the autonomous navigation agricultural machinery can effectively achieve the tasks of field sowing,fertilization,arable land and so on.However,when there are obstacles in the working environment,the autonomous navigation agricultural machinery can not identify and avoid the obstacles,which affects the normal operation of agricultural machinery.Therefore,based on machine vision technology,this paper designs obstacle avoidance path planning algorithm with autonomous navigation agricultural machinery as the test platform to help agricultural machinery to avoid obstacles in normal operation.The main research contents and conclusions are as follows:Firstly,a set of obstacle detection system based on binocular vision camera and lidar is designed.The design is mainly divided into hardware system and software system.The work of hardware system mainly includes:Design of portable power supply;realization of communication between binocular vision camera,lidar and industrial computer.The work of the software system mainly includes:improving the SIFT algorithm and realizing the first location and size measurement of obstacles;using lidar to realize the second location and size measurement of obstacles;based on the measurement data of the two,realizing the final location and size measurement of obstacles.Secondly,the intelligent upgrade of Dongfanghong SG250 tractor is carried out.The upgrade is mainly divided into three parts:navigation motion control system,navigation positioning system,navigation decision-making system.The work of the motion control system mainly includes:the measurement and control of the deflection angle of agricultural machinery.The work of the positioning system mainly includes:building a high-precision positioning module based on RTK-GPS;realizing the communication between GPS and industrial computer;completing the real-time positioning of agricultural machinery.The work of the navigation decision system mainly includes:the design of the deflection angle control loop of the agricultural machinery;the design of the position and attitude control loop of the agricultural machinery;the realization of the straight line tracking and the arc tracking of the autonomous navigation agricultural machinery;the integration of the deflection angle control loop and the position with attitude control loop of the agricultural machinery in the cascade control mode.Then,according to the location and size information of the obstacles determined above,the single obstacle avoidance path planning algorithm and the double/multiple obstacle avoidance path planning algorithm are proposed respectively.The latter algorithm is based on the former algorithm.According to the size of the collision free area,the obstacle avoidance of autonomous navigation agricultural machinery is realized from the left or right side of the obstacles.Finally,the whole autonomous navigation obstacle avoidance system is tested(1)The Improved SIFT algorithm is used for feature point detection test.Compared with the initial SIFT algorithm test data,the average number of feature point detection in four groups of tests of four graphs A,B,C,D is reduced by 260,and the detection time is reduced by 0.702s.When dealing with multiple images in a complex environment,the Improved SIFT algorithm can significantly improve the efficiency of the obstacle detection system.(2)When the test distance is 3.25m,the positioning errors of the left and right images are 0.06m and 0.08m respectively;when the test distance is 4m,the positioning errors of the left and right images are 0.34m and 0.20m respectively;when the test distance is 4.7m,the positioning errors of the left and right images are 0.84m and 1.77m respectively.With the increase of test distance,the positioning error of binocular vision camera also increases,and the positioning data is no longer reliable.When the measured size of the obstacle is 0.6m and the test distance is 3.25m,4m and 4.7m,the measurement error is stable within 0.05m,with small error and reliable measurement data.Under the same conditions,using lidar to locate obstacles,the average positioning error is 0.016m when the test distance is 3.25m,0.036m when the test distance is 4m,and 0.039m when the test distance is 4.7m.The lidar positioning error is small and stable.Therefore,the final positioning data of obstacles is provided by lidar,while the size measurement data is provided by binocular vision camera.(3)When a single obstacle is located at different positions and the traveling speed of agricultural machinery is 0.3m/s,the traveling path is 35%and 26%less than that of L algorithm,and the variance of agricultural machinery tracking error is 86%and 90%less;when the traveling speed is 0.5m/s,the traveling path is 38%and 22%less than that of L algorithm,and the square error of agricultural machinery tracking error is 99%and 76%less.When the speed of agricultural machinery is 0.3m/s and 0.5m/s,the path of double/multiple obstacle avoidance algorithm is 10.30m and 10.36m,which is only 0.09%and 0.06%higher than the planned path.The variance was 0.0067m2 and 0.0090m2 respectively.This algorithm has some advantages in driving path and planning path tracking stability.