FitTrack - Real-Time Exercise Posture Correction and Performance Monitoring Using Computer Vision

Abstract

Incorrectexercisepostureandtheabsenceofreal-timecorrectivefeedbackremainpersistentchallengesinhome-basedfitnes straining, particularly for individuals exercising without professional supervision.Poor exercise form not only diminishes workout effectiveness but substantiallyelevatestheriskofmusculoskeletalinjury. ThispaperpresentsFitTrack,avision-guidedreal-timefitnessmonitoringplatformthat performsautomatedpostureanalysisusinghumanposeestimationandafull-stackwebarchitecture. Thesystemcaptureslivevideoinputtodetectbodykeypoints,computejointangles,andevaluateexerciseformagainstpredefinedbiomechanicalthresholds.Basedonthisanalysis,itdelivers immediate visual and audio corrective feedback, enabling users to adjust their posture during active workout sessions. A finite state machine (FSM)governsrepetitioncountingand movementphasedetection ,whilesessionperformancemetrics—includingpostureaccuracyscoresand repetitioncounts—arepersistedinarelationaldatabase. AhybridAIcoachingengineintegratesGroqAIAPIstodelivercontextual,personalized coaching tips with a local JSON fallback ensuring uninterrupted guidance.Experimental evaluation demonstrates 90% posture detection accuracy and 95% repetition counting accuracy at 15–20 FPS on standard consumer hardware, requiring no wearable sensors or specialized equipment. FitTrackrepresentsanaccessible,low-latency,andtechnicallyrobustsolutionforimprovingworkoutsafety,trainingefficiency,andlong-term user engagement.

Introduction

FitTrack is an AI-powered, vision-based fitness monitoring platform that provides real-time exercise posture analysis, repetition counting, and personalized coaching without requiring wearable sensors or specialized hardware. Using computer vision and deep learning, it analyzes body movements through a standard camera, helping users perform exercises safely and effectively at home.

Problem

Current fitness applications primarily focus on metrics such as steps, calories burned, and heart rate, while largely ignoring exercise form and posture. Poor posture during exercises like squats, push-ups, and bicep curls can reduce workout effectiveness and increase the risk of injuries. Existing advanced monitoring systems often depend on expensive wearable sensors, EMG devices, or professional trainers, making them inaccessible to many users.

Motivation

The growing popularity of home-based and self-guided fitness routines has created a need for affordable systems that can:

• Monitor exercise posture in real time.
• Provide immediate corrective feedback.
• Improve workout safety and effectiveness.
• Operate using commonly available devices such as webcams and smartphones.

Solution

FitTrack addresses these challenges through a camera-only fitness monitoring system that:

• Uses pose estimation to track body movements.
• Computes joint angles to assess exercise form.
• Detects posture errors in real time.
• Counts repetitions accurately using a finite state machine (FSM).
• Provides visual and audio coaching feedback.
• Stores workout sessions for long-term performance analysis.

Key Technologies

• Human Pose Estimation: Uses MediaPipe BlazePose to detect 33 body landmarks from live video.
• Joint Angle Analysis: Calculates biomechanical joint angles using vector dot-product methods.
• Finite State Machine (FSM): Tracks movement phases and repetition counts.
• AI Coaching Engine: Delivers personalized exercise guidance using AI-generated and fallback coaching tips.
• Full-Stack Architecture: Built with:
  • React.js (Frontend)
  • Flask (Backend)
  • PostgreSQL (Database)

System Workflow

1. Capture live video through a camera.
2. Preprocess video frames.
3. Detect body landmarks using BlazePose.
4. Calculate joint angles relevant to the exercise.
5. Evaluate posture against predefined biomechanical rules.
6. Count repetitions using FSM-based movement tracking.
7. Generate real-time visual and audio feedback.
8. Store session metrics for future analysis.

Key Contributions

• Markerless fitness monitoring without wearable devices.
• Real-time posture correction using camera-based analysis.
• Accurate repetition counting and movement phase detection.
• AI-powered coaching with persistent workout analytics.
• Browser-accessible platform deployable on consumer hardware.

Performance Results

FitTrack demonstrated strong real-time performance:

• Posture Detection Accuracy: 90%
• Repetition Counting Accuracy: 95%
• Processing Speed: 15–20 FPS (frames per second) on standard consumer devices

Conclusion

ThispaperpresentedFitTrack,areal-timevision-guidedfitnessmonitoringplatformdesignedtoimproveexerciseposturesafety andprevent traininginjurieswithoutdependencyonwearablesensorsorspecializedhardware.ByintegratingMediaPipeBlazePose pose estimation, vector-based joint angle computation, FSM-driven repetition counting, and a hybrid Groq AI coaching engine within a full-stack web architecture, FitTrack delivers accurate, low-latency posture analysis and immediate corrective feedback accessiblefromanystandardcameraenableddevice.Experimentalevaluationconfirmed90%posturedetectionaccuracy,95%repetitioncountingaccuracy,andreal-timeperformanceat15–20FPSonconsumerhardware,establishingFitTrackasatechnically sound and practically accessible solution for home-based workout monitoring. Futuredevelopmentwillprioritizeexpandingtheexerciselibrarytoencompasscomplexmulti-jointmovementsincludingdeadlifts, overhead press, and lateral movements.Integration of personalized AI-driven adaptive training plans — adjusting exercise parametersbasedonhistoricalperformancetrajectories—willfurtherenhancecoachingvalue.PortingFitTracktoanativemobile applicationandincorporating smartwatch APIintegration forreal-timevitalsignmonitoring, includingheartrateandblood oxygen saturation, will extend the platform’s physiological monitoring scope and user engagement potential.

References

[1] Y.Chen et al., “Exercise assessment based on human pose estimation andrelative phase for real-time remote exercise system,”IEEEAccess,2025. [2] H.P.N.etal.,“Humanposeestimationforfitnessexercisemovementcorrection,”in2022InternationalElectronicsSymposium (IES), IEEE, 2022, pp. 1–6. [3] J.Wangetal.,“RGBD-basedreal-time3Dhumanposeestimationforfitnessassessment,”in2020IEEEInternational Conference on Bioinformatics and Biomedicine (BIBM), IEEE, 2020. [4] A.S.etal.,“Real-timehumanposeestimation:AMediaPipeandPythonapproachfor3Ddetectionandclassification,”in2023 IEEE International Conference, IEEE, 2023. [5] M.S.Al-awadhietal.,“AIvirtualgymtrainer:Usingdeeplearningforreal-timeexerciseguidanceandformcorrection,”in2024IEEEInternationalConference,IEEE,2024. [6] C.Lugaresietal.,“MediaPipe:Aframeworkforbuildingperceptionpipelines,”arXivpreprintarXiv:1906.08172,2019. [7] Z.Cao,T.Simon,S.Wei,andY.Sheikh,“Realtimemulti-person2Dposeestimationusingpartaffinityfields,”inProc.IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2017, pp. 7291–7299. [8] R.Dewangetal.,“Real-timeposturecorrectionofsquatexercise:Adeeplearningapproachforperformanceanalysisanderror correction,” IEEE Access, 2025. [9] NVIDIACorporation,“NVIDIADeepStreamSDK,”2024.[Online].Available:https://developer.nvidia.com/deepstream-sdk

Copyright

Copyright © 2026 Atharv Chougale, Jibran Attar, Mustafa Bhagat, Prof. Harsha Jain. This 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.