SYSTEM OF RECOGNITION OF AGGRESSIVE PHYSICAL BEHAVIOR ON VIDEO USING ARTIFICIAL INTELLIGENCE TECHNOLOGIES
There are no widely accepted solutions for detecting aggressive behavior, and many existing methods have low accuracy and are unable to recognize complex dynamic interactions. To ensure safety in public spaces and educational institutions, there is a growing need for specialized automated systems capable of preventing aggressive behavior in its early stages in real time. To address this problem, we propose an innovative system for the automatic detection, posture assessment, and classification of aggressive and defensive human behavior in video streams. The aim of this research is to develop an intelligent system capable of automatically determining a person’s psycho-emotional state from video recordings, based on the analysis of postures and behavioral patterns using computer vision methods. We employ a new multi-task architecture, in which we use YOLOv11-Pose for the first time to recognize aggressive actions. We created a unified trainable system that simultaneously performs detection, pose estimation, and classification. Video footage captured by surveillance cameras located in hallways, cafeterias, and other areas will be analyzed by the system to detect incidents and classify them. When recording visual signals of an assault, the recognition system identifies the participants in the incident, assigns them roles (“perpetrator” and “victim”), and records subsequent incidents involving them. The DataCollectorCV software tool was used for data annotation. The dataset consists of 1,034 manually annotated images. The developed system is capable of identifying multiple people simultaneously in real time based on their movements and poses and can serve as a valuable tool for analyzing psycho-emotional states. Quality testing confirmed the model’s reliability at 96%. Testing of the developed machine learning models for detecting physical bullying and aggression in video recordings yielded high results. The system can be used in educational institutions and in areas where video analysis is required.
Сведения
- Журнал
- Вестник КазАТК
- Опубликовано
- 2026 · Т. 4, № 145, С. 521-533
- Язык
- Английский
- ISSN
- 1609-1817 · 2790-5802
- Издатель
- JSC Academy of Logistics and Transport