FE-Fusion-VPR: Attention-Based Multi-Scale Network Architecture for Visual Place Recognition by Fusing Frames and Events

Traditional visual place recognition (VPR), usually using standard cameras, is easy to fail due glare or high-speed motion. By contrast, event cameras have the advantages of low latency, high temporal resolution, and dynamic range, which can deal with above issues. Nevertheless, are prone failure in motionless scenes, while still provide appearance information this case. Thus, exploiting complementarity effectively improve performance VPR algorithms. In paper, we propose FE-Fusion-VPR, an attention-based multi-scale network architecture for by fusing frames events. First, intensity frame volume fed into two-stream feature extraction shallow fusion. Next, three-scale features obtained through fusion aggregated three sub-descriptors VLAD layer. Finally, weight each sub-descriptor learned descriptor re-weighting obtain final refined descriptor. Experimental results show that our FE-Fusion-VPR outperforms existing frame-based, event-based fusion-based methods most cases on Brisbane-Event-VPR DDD20 datasets. a word, compared previous works, achieves new state-of-the-art (SOTA) datasets