Dynamic Obstacle Detection for Quadrotors with Event Cameras

Quadrotors based on traditional RGB cameras cannot effectively detect moving targets when facing high-speed moving obstacles, resulting in obstacle avoidance failure. Event cameras provide hardware support for capturing high-speed moving objects. This paper proposes a dynamic target detection method from the perspective of drones based on polar time difference surfaces. This method uses the spatial dislocation phenomenon of different polarities triggered by moving objects in the event stream to obtain the polar time difference surface, and finally realizes fast detection of dynamic targets through four-neighborhood depth-first search clustering. Experimental results show that this method obtains dynamic objects in the event stream with a delay of 5.92 milliseconds, which is 48.7% (5.63 milliseconds) less time than the existing motion target detection method, and improves the average detection recall rate, precision rate, and accuracy rate, proving the effectiveness of the method.