Research on human pose and action detection algorithm based on YOLOv8
Weizhong Cheng, Xuefei Julie Wang · 2025
Human pose and action detection technology has significant applications in areas such as smart security, sports monitoring, and human-computer interaction. The development of deep learning has driven the demand for real-time and efficient detection algorithms in this domain. Although existing human action detection algorithms have improved accuracy, their model complexity and computational demands hinder practical use in portable devices and edge computing platforms. To address this, the paper presents a lightweight human pose and action detection algorithm based on the YOLOv8 framework, designed to meet the high-precision, low-latency needs of edge computing and portable devices. In the feature extraction phase, GhostNet’s lightweight structure is used to reduce model parameters, while the CBAM attention mechanism, applied in the feature fusion phase, enhances the model's attention to critical features. This approach significantly decreases computational load while maintaining high accuracy. Experimental results indicate that the algorithm achieves a real-time detection rate exceeding 40 frames per second on edge computing platforms, demonstrating high accuracy and robustness across various human action scenarios. This offers strong technical support for human pose detection in smart security and wearable devices.