Research On Obstacle Avoidance Algorithm Based On Binocular Stereo Vision

In order to realize the autonomous navigation avoidance obstacle of the UAV in specific application scenarios.In this paper,the UAV obstacle avoidance algorithm based on binocular stereo vision is the research topic,focusing on binocular stereo vision technology,obstacle target detection algorithm integrating depth information and camera timestamp calibration problem with IMU.The contents are divided into Four parts shown as follows.Firstly,the design of the UAV obstacle ranging system is carried out.Research the principle and characteristics of mainstream UAV ranging methods(binocular vision,structured light and ultrasonic),design experimental comparison of various principles to represent sensor ranging accuracy,comprehensive principle characteristics and experimental results using binocular stereo vision sensor MYNT D100-IR-120 for obstacle ranging.Further research the principle of stereo matching of key steps in binocular ranging,and realize the transformation of obstacle position from pixel coordinate system to world coordinate system.Secondly,research the obstacles' s contour detection algorithm for UAV.The process of HSV color space conversion,morphological expansion corrosion and Canny operator edge detection based on color information detection algorithm is studied in detail.For the algorithm,the problem of non-robust detection is similar when the similar color obstacles overlap.A contour detection algorithm combining binocular depth information is used to perform depth segmentation on the image based on the detection result of the color information,and the experimental verification of the fusion depth information effectively improves the robustness of the detection of overlapping obstacles.Thirdly,research the UAV navigation algorithm based on VI-SLAM.From the four aspects of data preprocessing,IMU initialization,backend nonlinear optimization and loopback detection,research the principle of VI-SLAM algorithm for fusion visual inertia.The experimental results(VINS-mono and maplab)show that the latest algorithm solves the pose with certain error.Analyzes the above problems,the discovery is due to the time offset between the camera timestamp and the real camera trigger time.A camera timestamp calibration algorithm for joint visual inertia is proposed.The camera light is calculated by the zero-mean normalized cross-correlation algorithm.The similarity between flow and inertial navigation is used to solve the camera timestamp offset.The experimental simulation verifies the validity of the calibration algorithm and provides accurate navigation information for the autonomous flight of the drone.Finally,research the UAV path planning algorithm.Programming implements and research the A* algorithm process framework and mathematical principles.Pre experiments find that there is a problem that the search speed is slow and the path planning time is too long,which affects the flight safety of the drone.A method of storing the open list in combination with the data structure of the minimum binary heap is proposed.The simulation experiment is designed based on the optimized path planning algorithm.The path planning tasks can be efficiently completed for the three types of target points at different distances.The speed is increased by 60%,which verifies the effectiveness of the optimization algorithm.