Computer Vision Techniques in Violence Detection: An Overview

The current research paper discusses the importance of computer vision systems in every facet of life including security, health, commerce, and human-machine interaction. The paper shows that there are numerous computer vision methods of tracking human behavior and detecting violence in video. These techniques are convolutional neural networks (CNNs) to extract spatial features, deep time networks, including RNNs and LSTMs, to analyze the time sequence of events, transformer-based models, including ViViT and TimeSformer, to analyze long sequences of spatial and temporal data, and active optical flow methods to detect sudden changes between images. It is also used in the detection of violence cases, which is based on the posture of people and YOLO or object detection. Moreover, audio analysis and multimodal fusion are used to improve the algorithm of violence detection and supervisory self-learning, advanced spatiotemporal analysis, and matrix analysis are used to present more information about individual behavior. The pros and cons of computer vision-based violence detection algorithms are also compared.