Event-VPR: End-to-End Weakly Supervised Deep Network Architecture for Visual Place Recognition Using Event-Based Vision Sensor

Traditional visual place recognition (VPR) methods generally use frame-based cameras, which will easily fail due to rapid illumination changes or fast motion. To overcome this, we propose an end-to-end VPR network using event can achieve good performance in challenging environments (e.g., large-scale driving scenes). The key idea of the proposed algorithm is first characterize streams with EST voxel grid representation, then extract features a deep residual network, and, finally, aggregate improved VLAD realize streams. verify effectiveness algorithm, on event-based datasets (MVSEC, DDD17, and Brisbane-Event-VPR) synthetic (Oxford RobotCar CARLA), analyze our method sequences, including cross-weather, cross-season, changing scenes, then, compare state-of-the-art (Ensemble-Event-VPR) prove its advantages. Experimental results show that better than ensemble scheme scenarios. best knowledge, for task, this weakly supervised architecture directly processes stream data.