Tsagaantsooj Batzaya, Eunbi Cho, Jeong-Hyeon Chang, Kwanghoon Pio Kim, Journal of Internet Computing and Services, Vol. 26, No. 6, pp. 83-92, Dec. 2025
10.7472/jksii.2025.26.6.83, Full Text:  HTML
Keywords: CCTV, Risk Behavior, Object Detection, Dynamic interaction, Weapon Analysis

Abstract

Real-time identification of weapon-related dangerous behaviors in CCTV footage is essential for public safety, yet most existing systems rely on static object recognition without incorporating behavioral dynamics. This study applies YOLOv8-based object detection and tracking to 174,372 frames from the film The Outlaws to analyze risky behaviors defined by human–weapon distance, movement speed, and positional interaction. Results show that humans consistently appeared as central actors in all risky frames, and each weapon type exhibited distinct behavioral patterns: baseballbats produced sustained mid-range attacks, knives demonstrated slow but deliberate close-range approaches, and scissors generated rapid and impulsive short-range movements. Peaks in human counts and the presence of a single weapon further served as indicators of elevated risk. These findings demonstrate that accurate identification of dangerous behaviors requires integrating object presence with motion features and temporal co-occurrence. The proposed analytical framework offers a practical basis for enhancing real-time CCTV threat detection and supporting public safety systems.

Statistics
Cite this article