Research On Obstacle Detection And Obstacle Avoidance Strategy For Unmanned Aerial Vehicle

In order to ensure the flight safety of unmanned aerial vehicle(UAV),especially in the unknown environment when the industrial UAV operates beyond the horizon,the obstacle detection and obstacle avoidance strategy is studied.Obstacle detection is divided into two aspects: obstacle recognition and obstacle distance measurement.In the aspect of obstacle recognition,the edge extraction algorithm is used to extract the contour of obstacles in the image,then the obstacles are selected by rectangular boxes,and the position,shape and size of obstacles are determined by replacing irregular true contours with regular rectangular contours.Original Local Binary Patterns(LBP)operator is improved by using the theory of fuzzy set in Hue,Saturation,Value(HSV)color space to realize the extraction of obstacle contour.The problem that the difference of original LBP operator can not be quantified and has no rotation invariance is solved.In the case of weak light or noise contained,it performs well and has a good real-time performance.On Jetson Tk1 embedded development board,it takes an average of 75 ms to process an image with 720 P resolution,which can meet the real-time requirements of UAV obstacle avoidance.In the aspect of obstacle recognition,binocular camera and millimeter wave radars installed in different directions are used to measure obstacles' distance.The distance measurement of obstacles can be realized by using the depth map generated by binocular camera and the rectangular contour of obstacles.In order to solve the problem that the measurement frequency of binocular camera is only 10 Hz,millimeter wave radar whose measurement frequency is 50 Hz is introduced to assist ranging.The center area of binocular camera depth map and forward millimeter wave radar use multisensor data fusion algorithm for fusion ranging.By expanding the dimension of state matrix and measurement matrix,it converts the asynchronous fusion problem of measurement data with different sampling rates into synchronous fusion problem.The measurement data has higher accuracy than single sensor.By calibrating the center area of depth map and the measuring angle of forward millimeter wave radar,the range of fusion can be ensured to measure the obstacles appearing in the forward flight path of UAV.The results of the fusion ranging and the results of using the rectangular contour of obstacles in depth map take a smaller value as the forward obstacle distance to ensure flight safety.Millimeter-wave radar measurements in other directions are used to assist in judging whether the obstacle avoidance process is completed and whether abnormal conditions occur.In the aspect of obstacle avoidance strategy,computational geometry is used to determine whether obstacles in the image will hinder UAV forward flight and calculate the optimal direction and the speed of UAV to avoid obstacles.The forward flight speed of UAV is limited according to the obstacle distance to ensure that UAV will not collide with obstacle.Simplifying the multi-obstacle case into a single obstacle case can reduce the complexity.The flight experiment verifies the effectiveness of the obstacle avoidance method in this paper.The UAV can safely avoid obstacles in the best direction and return to the preset flight trajectory.