Urban Street Cleanliness Assessment by Mobile Object Detection

Road garbage recognition and road cleanliness accessment play an important role in road informatization management system, which can enhance the efficiency of urban public health management. However, no public dataset can be used and there are scare amounts of literature in this work. This paper proposes an end-to-end re-parameterized and attention-based Yolo network (RAYolo) and a dataset for road garbage recognition. This model consists of 1) a re-parameterized structural backbone, which can boost the inference speed, 2) channel attention mechanism, which is used to improve the channel and position relevance. Compared with the baseline and other existing methods, the proposed method can do a better work. The effectiveness of proposed method outperforms 1.2% than the baseline at most in mAP.5:.95, and the inference time and params can even reduce 45.2% at most. And after the recognition network, a novel, simple and accurate method is established to assess road cleanliness by combining the results from garbage recognition. Experiments results show that the proposed road garbage detection network and cleanliness assessment methods can provide accurate road cleanliness information for road informatization management system.