Enhancing Object Detection Accuracy Through RGB and Event Fusion in Motion Blurred Images

초록

Deep learning-based object detection exhibits impressive levels of accuracy. However, its performance significantly degrades when motion blur occurs in images captured with an RGB camera. In contrast, an event camera is a sensor that detects brightness changes for each pixel, offering advantages such as high temporal resolution and low latency. It has been recently applied to various computer vision tasks. In this paper, we propose a fused image that combines RGB and event data, leveraging the advantages of the event camera to enhance detection accuracy in motion-blurred images. No additional operations are required for deblurring. Thus, it incurs minimal computational overhead and can be used without modifying the existing object detection model structure. © 2024 IEEE.