Continuous Histogram for Event-Based Vision Camera Systems

In object detection using event-based vision cameras, most current methods rely on recurrent layers. However, training recurrent layers requires continuous time event data, which means more data and longer training times compared to RGB image training. In addition, most low-cost edge devices do not support recurrent layers, making it harder to use these methods widely. To solve this issue, we propose a continuous histogram that combines surface and event histogram. We also introduce an edge enhancement method within the continuous histogram. Our approach enables efficient object detection using event data without relying on recurrent layers, achieving 56.7% mAP in the Gen1 detection test dataset and 51.2% mAP in the 1Mpx detection test dataset, showing competitive performance compared to state-of-the-art methods.