This paper presents the research being done at Cal Poly Portion on the development of an obstacle avoidance system for use on small, fixed-wing UAVs. Computer based vision is used in conjunction with an automatic flight control system to detect and avoid obstacles in the UAV's flight path. The computer vision utilizes video captured from one tail-mounted camera. The frames of the video stream are inputted to the vision algorithm, where a hue saturation value filter developed from the OpenCV library separates objects from the background and determines which objects are of interest. Using a two-frame differential optical flow method, called the Lucas-Kanade method, pixel locations of objects of interest in a captured frame are compared to the same objects in the prior frame, thus creating a velocity vector and allowing future predictions of vectors to be made. The velocity vector values are outputted to the flight controller, which compares them to predetermined threshold values in order to make the correct avoidance decision.
展开▼
