Identifying Loitering Behavior with Trajectory Analysis

Loitering — remaining in a public area for an extended period — is often associated with suspicious activity and public safety concerns. Research on loitering detection has lacked standardization, making it difficult to generalize detection methods across scenarios.

In this work, we provide a clear definition of loitering and introduce a new dataset with annotated loitering behaviors drawn from the Long-term Thermal Drift Dataset, which respects privacy standards. We utilize trajectory analysis methods to identify loitering by quantifying parameters such as movement directionality, pace, and dwell time.

Key contributions:

• Standardized definition of loitering behavior for computer vision research
• New annotated dataset from thermal surveillance respecting privacy
• Geometric motion descriptors for robust trajectory-based detection
• Open-source dataset and code

Dataset and code: github.com/johnnynunez/RS-WACV24_Loitering

Recommended citation: Johnny Núñez, Zenjie Li, Sergio Escalera, Kamal Nasrollahi. (2024). "Identifying Loitering Behavior with Trajectory Analysis." Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). 251-259.
Download Paper | Download Slides