The increasing rate of street-level crimes and accidents in urban environments necessitates the development of intelligent and automated surveillance systems. Traditional monitoring approaches rely heavily on human observation, making them inefficient and prone to errors. This paper presents a behaviour-based crime detection framework integrated with an autonomous surveillance system using real-time object detection and tracking. The proposed system utilizes the YOLOv8 deep learning model to detect objects such as persons and vehicles from video input, followed by object tracking to maintain identity across frames. Behavioural analysis is performed using spatial and temporal features, including movement speed and interaction between objects. Suspicious activities are identified based on predefined rules, such as sudden speed changes and close proximity interactions. Furthermore, a decision module generates control signals that can assist autonomous drone systems in tracking suspects. The system operates in real time and does not require additional training, making it efficient and scalable for practical surveillance applications. Experimental results demonstrate that the proposed framework effectively detects abnormal activities and enhances situational awareness in dynamic environments.
Panchetti et al. (2026) studied this question.