Enhancing Event-based Object Detection with Monocular Normal Maps
Researchers have developed NRE-Net, a novel trimodal framework designed to enhance object detection for autonomous driving systems, particularly in challenging lighting conditions. This new approach integrates surface normal maps derived from RGB images to provide geometric constraints, which are crucial for overcoming misleading event signals from event cameras. The framework's Adaptive Dual-stream Fusion Module and Event-modality Aware Fusion Module effectively combine structural priors, appearance context, and dynamic event data, leading to significant performance improvements over existing methods. AI
IMPACT This research could improve the reliability of autonomous driving systems by enhancing object detection accuracy in adverse lighting conditions.