Occlusion-guided multi-modal fusion for vehicle-infrastructure cooperative 3D object detection

Huazhen Chu, Haizhuang Liu, Junbao Zhuo, Jiansheng Chen, Huimin Ma

Abstract: Highlights•Innovative Occlusion Confidence-Aware Communication. We introduce a technique to quantify occlusion levels, allowing for smart data transmission prioritization. This method optimizes bandwidth by concentrating on the most critically affected areas.•Strategic Foreground Feature Fusion. Fusion2comm uses selective fusion to combine key foreground features, optimizing bandwidth by cutting down on background data transmission.•Performance. Experimental results on the DAIR-V2X dataset show that our method significantly improves detection accuracy and communication efficiency.