Publication: Vision-based human violence detection using yolov8 object detection algorithm
Loading...
Date
2024-07
Authors
Muhammad Arsyad bin Rahman Yusof
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Despite advancements in violence detection technologies, existing models face challenges in real-time processing and generalization across diverse scenarios,
necessitating the development of a more efficient and accurate system. The primary objective is to develop an efficient deep learning model for real-time violence
detection using the YOLOv8 object detection framework, thereby improving the effectiveness and reliability of surveillance systems in diverse and dynamic real-world environments. The methodology involves compiling a diverse dataset of violent and non-violent interactions from multiple online repositories, employing data augmentation techniques, and training the YOLOv8 model using transfer learning. The YOLOv8 model achieved a precision of 80.7% recall of 74.2%, an F1 score of 77.42%, and accuracy of 82.5% demonstrating superior performance in detecting violent and able to process in real-time. The model's robustness and generalization capabilities were validated across various test scenarios. This research successfully developed a effective and efficient vision-based violence detection system using the YOLOv8 model, offering significant potential for application in real-world security and surveillance systems.