Safety Net Detection by Optic Flow Processing

Drone navigation is an area of study that is receiving more and more attention. Obstacle detection techniques and autonomous guidance are continuously improving, but some types of obstacles are still very difficult to detect with current methods. Safety nets used to separate and secure 2 contiguous spaces are indeed very difficult to detect by Lidar and by image processing based on pattern recognition. The method we propose here separates the Optical Flow detections to identify the presence of a safety net: i) by using the norm of their vector, ii) by matching them to a regression defining a plane (safety net or wall). Our results show that the proposed method detects a net in front of a wall with very few false positives, thanks to a small displacement (at most 5%). Moreover, the distance estimation between the net and the wall as well as the distance between the net and the drone can be estimated with at most 20% error in the worst cases.