Research Of Obstacle Avoidance Based On Monocular Vision For Unmanned Aerial Vehicles

Unmanned aerial vehicles(UAVs) are widely used in all aspects in human life and played a major role in tasks such as searching and rescuing, environment monitoring and so on. Obstacle detection and avoidance is a challenging problem for UAVs. The major work of this paper is to deal with the problem using monocular vision in order to avoid obstacles when UAVs on its flight. Let the UAVs can autonomous flight without prior setting its flight path.Monocular camera as a sensor to obtain the information for UAV, the whole process of the obstacle avoidance are divided into five parts: image preprocessing, feature detection, feature matching, depth information calculation and obstacles avoidance method. In terms of feature point detection, an improved SIFT feature point detection algorithm is proposed. In this algorithm, segment the original feature points description region by concentric circles, and count the gradient value of all the pixels in each circle. For the pixels that the circles cross over,merge it to neighboring circles by half of the gradient value. Then dimension of the feature descriptor described from 128 d down to 40 d,while the range from 16 ×16 up to 20 ×20. In terms of computing depth information, an image template is generated by the neighborhood of the detected feature point. The relationship between the feature’s scale is calculated with the temple matching method.The formula of distance information is deduced by similarity theorem. The depth information is computed according to the distance formula and the speed of the UAV. The depth image of scenario is obtained after convex hull for all the feature points. In terms of obstacles avoidance method, a new layered-graph is acquired with the depth image by distance of obstacles avoidance. Select a region where include the security zones that the UAV can crossing. lastly, figure out the location of the security zones by the means of block search in the region and calculate the distance of obstacle avoidance for the UAV.According to the experience result, the improved SIFT algorithm can not only ensure the accuracy of the original algorithm, but also reduce thecomputing time. The simplified depth map calculated with proportion of size can meet the requirements of obstacle avoidance. The cost time of the method based on regional block search can be greatly reduced than search the whole image, and greatly improve the real-time performance of obstacle avoidance. The method based on monocular vision for UAV obstacle avoidance has feasibility and research significance.