Abstract:Visual system is an important part of autonomous unmanned system. The raindrops attached to the surface of the camera lens will produce artifacts in the image, resulting in the degradation of image quality, which will significantly affect the performance of the visual system. The detection and removal methods of attached raindrops in recent years were comprehensively and deeply researched. The essence of the problem was condensed, and the existing raindrop imaging models were summarized. Different technical methods were sorted out from three directions:model-based, data-driven and camera system-based, then the development of deep network model was summarized from two aspects of network architecture and loss function. The existing datasets of attached raindrops were summarized, and the performance of some algorithms was compared through the experimental results. The main problems in the task of raindrop detection and removal were discussed, and the possible development direction in this field was prospected.