*Result*: A Real-Time System for Athlete Pose Analysis and Feedback Based on Machine Learning.

*Accurate and real-time analysis of athlete posture is an important topic in sports-related image processing. In practical training environments, pose analysis systems are expected to operate under strict real-time constraints while remaining robust to occlusion, motion blur, and scene variation. However, existing approaches face several limitations. 0054hree-dimensional pose estimation often depends on large amounts of annotated data, which are expensive and difficult to obtain. Lightweight models designed for real-time inference tend to sacrifice spatiotemporal feature representation, leading to reduced accuracy. In addition, current feedback mechanisms are usually loosely connected to the underlying pose features and therefore provide limited diagnostic value. To address these issues, a real-time pose analysis and feedback framework based on self-supervised spatiotemporal optimization is presented. The system adopts a three-stage architecture consisting of a lightweight image feature extraction and two-dimensional keypoint detection module, a dual-path spatiotemporal feature refinement module, and a sequence-based feedback generation module. The refinement stage combines adaptive graph convolution for skeletal topology modeling with a lightweight spatiotemporal Transformer for learning temporal image features. Temporal coherence across video frames is exploited to construct self-supervised constraints for three-dimensional pose learning without manual annotations. Pose sequences are further matched with standard motion templates using dynamic time warping, and the resulting deviations are translated into structured feedback. The proposed framework reduces the dependence on annotated data, maintains real-time performance on edge devices, and provides interpretable feedback linked directly to pose deviations. Experimental results demonstrate that the system achieves a balanced trade-off between efficiency, accuracy, and practical usability in real training scenarios. [ABSTRACT FROM AUTHOR]

Copyright of Traitement du Signal is the property of International Information & Engineering Technology Association (IIETA) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)*