Identifying Loitering Behavior With Trajectory Analysis

Johnny Núñez, Zenjie Li, Sergio Escalera, Kamal Nasrollahi; Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) Workshops, 2024, pp. 251-259

The act of remaining in a public area for an extended period is commonly referred to as Loitering, and it is often viewed as suspicious activity with regard to public safety. The research landscape on loitering detection is diverse, featuring various definitions and methodologies. This lack of standardization in defining loitering hamper the generalizability of detection methods. Our work focuses on providing a clear definition of loitering and detecting it through trajectory analysis. We enrich the field of loitering detection research by introducing a dataset with annotated loitering behaviors. Our contribution is to annotate loitering behavior in the Long-term Thermal Drift Dataset, which already complies with privacy standards. The dataset features a variety of loitering behaviors observed through a real-world thermal surveillance camera across different environmental scenarios. To identify loitering behavior, we employ trajectory analysis methods. These methods quantify parameters such as movement directionality, pace, and dwell time, providing fundamental aspects for loitering detection studies. The dataset and the code are available on https://github.com/johnnynunez/RS-WACV24_Loitering.