A multisource human pose stability recognition method based on a seven-point structural mode

In public environments, human behavior recognition is essential for ensuring safety. Moving beyond conventional behavior categorization, it is increasingly important to explore the fundamental theories underlying human postural stability. This study proposes a human stability recognition method based on physical dynamic modeling, emphasizing the high degree of compatibility between the stability representation model and physical theoretical analysis. A seven-point structural model is constructed, which comprehensively encompasses the core anatomical structures required for effective stability analysis. Based on screw theory, the kinematic characteristics of critical joints are modeled, and the dynamic equations of the system are derived. A multidimensional set of physically meaningful features is subsequently extracted to characterize the human stability state. Experimental results demonstrate that the proposed method achieves a recognition accuracy of 96.1% for common instability behaviors, with a per-frame processing time improvement of approximately 50%. This approach offers strong physical interpretability and generalizability, making it well-suited for real-time stability assessment in various safety-critical scenarios.