Monocular Vision-Based Dynamic Moving Obstacles Detection and Avoidance

In this paper, we proposed an UAV system for obstacle avoidance depending of a ground station processing based on monocular vision. To accomplish detection, tracking, proximity estimation and dynamic obstacles avoidance that are approaching the UAV, a series of methods and techniques are implemented. To detect movement, frame differentiation was applied to consecutive frames, once the moving object is detected, we detect Shi-Tomasi feature points to track the object using optical flow method Lucas-Kanade. To make possible proximity estimation, a linear regression method based on the area covered by the object was used. A fuzzy logic controller was designed to avoid the moving object, we considered the approach rate and the area of the object to control UAV's position in relation to the object as fuzzy inputs.