Real-Time Human Pose Estimation and Violence Detection

Devika Suresh, Jogimol Joseph

Abstract: Real-time human activity recognition and violence detection have become essential for modern surveillance and safety applications. This paper presents an AI-based system for human pose estimation and activity recognition using deep learning techniques. The proposed system detects humans in video frames and extracts skeletal keypoints using YOLOv11. These keypoints are analyzed over time using a temporal model such as LSTM or Transformer to recognize activities including walking, running, and fighting. Violence detection is achieved by identifying abnormal motion patterns, aggressive interactions, and rapid body movements. Unlike traditional surveillance systems that rely on manual monitoring or basic motion detection, the proposed approach provides improved accuracy and real-time performance. The system is implemented using Python and OpenCV and is designed to be scalable and efficient without requiring specialized hardware. Experimental results demonstrate the effectiveness of the system in accurately detecting human activities and identifying violent behavior, making it suitable for applications in public safety and intelligent surveillance systems.

Keywords: Real-time human activity recognition and violence detection have become essential for modern surveillance and safety applications. This paper presents an AI-based system for human pose estimation and activity recognition using deep learning techniques. The proposed system detects humans in video frames and extracts skeletal keypoints using YOLOv11. These keypoints are analyzed over time using a temporal model such as LSTM or Transformer to recognize activities including walking, running, and fighting. Violence detection is achieved by identifying abnormal motion patterns, aggressive interactions, and rapid body movements. Unlike traditional surveillance systems that rely on manual monitoring or basic motion detection, the proposed approach provides improved accuracy and real-time performance. The system is implemented using Python and OpenCV and is designed to be scalable and efficient without requiring specialized hardware. Experimental results demonstrate the effectiveness of the system in accurately detecting human activities and identifying violent behavior, making it suitable for applications in public safety and intelligent surveillance systems.

How to Cite?: Devika Suresh, Jogimol Joseph, "Real-Time Human Pose Estimation and Violence Detection", Volume 15 Issue 4, April 2026, International Journal of Science and Research (IJSR), Pages: 1589-1594, https://www.ijsr.net/getabstract.php?paperid=SR26424151137, DOI: https://dx.dx.doi.org/10.21275/SR26424151137

Download Citation: APA | MLA | BibTeX | EndNote | RefMan