Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{205417,
        author = {Atharv Chougale and Jibran Attar and Mustafa Bhagat and Harsha Jain},
        title = {Real-Time Biomechanical Fatigue Modeling and Injury Risk Prediction Using Vision-Based Human Pose Estimation},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {1},
        pages = {8956-8965},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=205417},
        abstract = {Vision-based exercise monitoring systems commonly track joint angles and repetition counts without modeling the physiological fatigue state associated with musculoskeletal overuse risk. This paper presents a real-time systems framework—to the best of our knowledge, the first camera-only exercise monitor to couple markerless pose estimation with a per-muscle thermodynamic fatigue model and a grammar-constrained speech interface within a single integrated pipeline. For each active muscle group the system estimates motor-unit recruitment, mechanical tension, tissue temperature, and cumulative fatigue-indicator scores through coupled difference equations whose parameters are grounded in established exercise physiology literature on metabolic heat production [11], Henneman recruitment ordering [13], and two-compartment thermal muscle models [14]. These estimates serve as physically motivated heuristic proxies for fatigue-associated injury risk indicators and are not intended as clinically validated physiological measurements. Posture correctness is evaluated using a scale-invariant elbow-drift metric, and repetitions are counted by a two-threshold finite-state machine that eliminates double-counting artifacts. 
A grammar-constrained Bayesian de-coder restricted to a 17-phrase domain vocabulary holds audio-processing load at approximately 1% CPU utilization, enabling the vision pipeline to sustain 15–20 fps on a consumer-grade laptop. Out-of-vocabulary rejection evaluated against 500 high-exertion acoustic events produced zero false command triggers.},
        keywords = {Biomechanical modeling, human pose estimation, fatigue analytics, motor-unit recruitment, Bayesian speech de-coding, real-time systems.},
        month = {June},
        }