Adaptive spatio-temporal clustering for small object detection using event-based camera

Abstract

Event-based camera outputs asynchronous temporal and spatial pulses with high temporal resolution and dynamic range, and is capable to image high-speed, extreme light conditions. In this paper, we propose an adaptive point cloud spatiotemporal clustering method to detect small moving objects in event streams. Specifically, our method first transforms the raw event stream into space-time point clouds. Due to the high frame rate of event cameras, moving small objects produce groups of points with high density in the point cloud. Then, our method traverses the point cloud to separate points with an adaptive threshold to detect the moving objects. We test our method on several video sequences with challenging scenarios captured by event camera. Experimental results show that our method can effectively detect the moving objects with promising accuracy.