High-temporal-resolution event-based vehicle detection and tracking

Event-based vision has been rapidly growing in recent years justified by the unique characteristics it presents such as its high temporal resolutions (~1us), dynamic range (>120dB), and output latency of only a few microseconds. This work further explores hybrid, multi-modal, approach for object detection tracking that leverages state-of-the-art frame-based detectors complemented hand-crafted event-based methods to improve overall performance with minimal computational overhead. The presented include bounding box (BB) refinement improves precision resulting BBs, well continuous method, recover missed detections generate inter-frame enable high-temporal-resolution output. advantages these are quantitatively verified an ablation study using higher order accuracy (HOTA) metric. Results show significant gains resembled improvement HOTA from 56.6%, frames, 64.1% 64.9%, event edge-based mask configurations combined two proposed, at baseline framerate 24Hz. Likewise, incorporating same improved 52.5% 63.1%, 51.3% 60.2% rate 384Hz. Finally, validation experiment is conducted analyze real-world single-object high-speed LiDAR. Empirical evidence shows our approaches provide compared 24Hz rates up 500Hz.