Obstacle Avoidance for Unmanned Aerial Vehicles

Obstacle avoidance for unmanned aerial vehicles (UAVs) in GPS-denied environments is swiftly gaining popularity in the research community. Our proposed method relies solely on the scene structure that the frontal monocular camera of a UAV acquires and performs a lightweight predictive operation for the next course of maneuver. The object's relative size on the image plane continues to increase as the UAV moves toward it. Assuming that the object contains many scale-invariant visual keypoints, we can use the increase in the Euclidean distance for each, from the centroid of all the keypoints in successive frames, to measure the depth of the obstacle. In this regard, we compute some scale-invariant keypoints in the field of view (FoV) and monitor the Euclidean distance expansion with the movement of the UAV to estimate the obstacle's position. We conducted several rounds of experiments with varying obstacles to reveal the accuracy of our proposed system. © 2012 IEEE.