Towards Versatile Pedestrian Detector with Multisensory-Matching and Multispectral Recalling Memory

Recently, automated surveillance cameras can change a visible sensor and thermal for all-day operation. However, existing single-modal pedestrian detectors mainly focus on detecting pedestrians in only one specific modality (i.e., or thermal), so they cannot cope with other modal inputs. In addition, recent multispectral have shown remarkable performance by adopting modalities, but also limitations practical applications (e.g., different Field-of-View (FoV) frame rate). this paper, we introduce versatile detector that shows robust detection any single modality. We propose multisensory-matching contrastive loss to reduce the difference between visual representation of modalities. Moreover, modality, design Multispectral Recalling (MSR) Memory. The MSR Memory enhances features recalling To guide store contexts, loss. It enables encode more discriminative input believe our method is step forward be applied variety real-world applications. comprehensive experimental results verify effectiveness proposed method.