FramePoseNet: A Multi-Scale and Temporal Deep Learning Model for Real-Time Human Pose Detection in Highly Dynamic Motion Scenarios

Human pose detection in highly dynamic motion scenarios is a key technology in the fields of action analysis and motion monitoring, but existing methods still face challenges in balancing between real-time and detection accuracy. To address this problem, this paper proposes FramePoseNet, a deep learning model that combines multi-scale feature extraction, temporal information modeling and dynamic adaptive computing. FramePoseNet’s unique advantages lie in its multi-scale convolutional module for capturing pose details at different scales, temporal information processing module for ensuring motion coherence, and dynamic response adaptive module for optimizing computational efficiency. These innovations enable FramePoseNet to achieve superior detection accuracy and real-time performance, particularly in highly dynamic motion scenarios. Experimental results on the COCO Keypoints dataset demonstrate that FramePoseNet outperforms existing mainstream models in mean keypoint accuracy (mAP) and real-time detection, offering a reliable and efficient solution for real-time pose detection in complex dynamic environments. The proposed model has wide application potential in areas such as sports analysis and real-time action monitoring.