ONEWO: A Dataset of Anomaly and Object Detection for Community Surveillance

In the era of smart cities, the role of surveillance systems in ensuring community safety and environmental sustainability has become increasingly significant. Traditional surveillance methods, reliant on human monitoring, often exhibit deficiencies in terms of speed, accuracy, and scalability. However, to the best of our knowledge, there is no related dataset on this research field. This paper introduces a novel approach to address these challenges, leveraging the power of computer vision technology to enhance surveillance capabilities. To bridge the gap between algorithms research and application, in this paper, we propose the task and dataset, and highlight the necessity of camera grading. ONEWO comprises 18,307 images from 2,863 cameras, which covers 20 main objects of interest for city management, providing annotation of bounding boxes and segmentation masks of 121,342 instances. Experiments demonstrate that our best baseline model achieves mAP
50
of 77.5 for the entire test set. This work aims at contributing to the development of more efficient, accurate, and scalable surveillance systems, thereby promoting community safety and environmental sustainability in smart cities.