Real-Time Injury Risk Assessment in Athletes Based on Relative Joint Angles

Preventing running injuries is critical for track and field athletes. This work presents a novel machine learning methodology to automatically detect injury risk from videos by analyzing biomechanical technique. Joint angles from anatomical pose estimates serve as descriptive features input to ML classifiers. Statistical tests verified significant distribution differences in angles between injured and non-injured states, evidencing utility as injury markers. Multi-frame analysis integrating temporal context attained 0.735 accuracy on held-out test data using SVM, surpassing single-frame approaches. The scoring function showed no significant statistical drift from training to test, providing evidence of strong generalization. By automating nuanced pose pattern quantification imperceptible to the naked eye, the proposed video screening enables early at-risk athlete identification for targeted preventative intervention before injuries develop.