Many existing intelligent CCTV systems rely on rule-based models, leading to frequent errors due to environmental factors, high resource burdens in large-scale environments, and inefficient manual configuration processes. Therefore, to address these issues, this study proposes an anomaly detection model that integrates deep learning-based object tracking technology with a hybrid clustering technique based on distance and density. The proposed method utilizes YOLO and DeepSORT algorithms to perform precise object detection and tracking, and the acquired spatiotemporal coordinate data undergoes a two-stage clustering process. First, the K-Means algorithm is employed to calculate the Cluster Centroid Distance to define primary movement paths and perform an initial screening of geometric deviations. Subsequently, the DBSCAN algorithm is applied to analyze spatiotemporal density, thereby finally detecting unstructured, rare anomalous trajectories. Experimental results confirm that the proposed model detects unstructured abnormal behaviors (such as S-shaped trajectories) more efficiently than simple distance-based analysis and significantly improves management convenience and detection accuracy through automated configuration.
Seo et al. (Tue,) studied this question.