Research And Design Of Obstacle Avoidance Based On Binocular Vision

With the wide application of obstacle avoidance technology in military fields,scientific detection,traffic control,industrial manufacturing,medical services and other fields,various machines with autonomous navigation and obstacle avoidance capabilities have been used to varying degrees.Thereby replacing some daily production activities of human beings and bringing many conveniences to people's lives.The prerequisite for the obstacle avoidance system in obstacle avoidance path planning is to be able to obtain the position information of obstacles in its forward direction in real time,which is the key to the machine's autonomous obstacle avoidance.In this paper,an obstacle avoidance system is designed based on binocular vision theory,and the mobile machine's orientation of environmental obstacles is effectively detected.main tasks as follows:(1)The overall design of the obstacle avoidance system.In this article,first of all,the system's process design and function module design are completed,and the obstacle avoidance algorithm software and hardware platform are built.Then,we has verified the feasibility of the system's autonomous obstacle avoidance in a complex environment,and realize the design of each functional module.(2)Implementation of obtaining depth image information.Through the research and implementation of the binocular vision theory,depth image information can be obtained.First,through the Matlab camera calibration tool,we have achieved binocular camera calibration,so as to obtain the internal and external parameters of the camera calibration.Then,we use OpenCV's epipolar correction function to correct the image based on these calibrated parameters.Finally,the OpenCV SGBM stereo matching function is used to obtain the depth image according to the corrected image.In this process,we applied the image pyramid method to improve the efficiency of binocular visual stereo matching to obtain depth images.While maintaining the original stereo matching effect,the efficiency of the algorithm is improved as much as possible.In the experiments in this paper,although the mismatch rate of the stereo matching algorithm hasincreased by 1.85%,the algorithm efficiency has increased by 10 times.We analyzed the experimental results of the indoor object distance measurement.The maximum error of the object distance measurement is ±1.49%,and its accuracy can reach 98.9% within the 3 meter measurement range.(3)Obstacle detection completed.Based on the knowledge of obstacle-related theoretical models,we use a combination of depth information and color information to segment obstacles.Through the fusion of super-pixel segmented images,the influence of noise,background and ground interference factors is effectively eliminated,thereby improving the recognition accuracy of obstacle data.On the PC platform,we completed the measurement of real data of obstacles.By comparing the obstacle detection results,it can be obtained that the position information of the obstacle obtained by combining the superpixel segmentation method is more accurate.(4)Research on local path planning algorithm for dynamic obstacle avoidance.According to the data information of obstacles,we studied the improved method of artificial potential field method,which improved the ability to reach the target point safely in path planning and overcome the problem of local minimum point in the algorithm.The simulation experiment results of the algorithm show that the improved algorithm achieves the purpose of path planning,and the arrival rate of obstacle avoidance is 99.07%.