Looming object detection with event-based cameras
Lethbridge, Alta. : University of Lethbridge, Dept. of Mathematics and Computer Sciences
We present a looming object detection method for event-based cameras. Event-based cameras detect events asynchronously which eliminates the unnecessary computation required for the conventional frame-based cameras. There are two main parts of this method. In the first part, we develop an event-based optical flow algorithm. The algorithm is based on Reichardt motion detectors inspired by the fly visual system and has a very low computational requirement for each event received from the event-based camera. In the second part, we develop an algorithm to detect looming objects using the output from the first algorithm. This proposed method is only sensitive to significant log-luminance changes, which results in low energy consumption. We have performed several experiments with our method using the Davis Dynamic Vision Sensor (DVS) which is an event-based camera. Experimental results show that our event-based looming detection algorithm accurately detects looming objects in all cases when there is a single object moving in the scene. It also does not report looming when no objects are looming. Our algorithm is fast and operates in real-time, requiring only microseconds to process each event.
Communication and technology , Computer systems , event-based cameras , looming object detection