International Journal of Innovative Research in Computer and Communication Engineering
ISSN Approved Journal | Impact factor: 8.771 | ESTD: 2013 | Follows UGC CARE Journal Norms and Guidelines
| Monthly, Peer-Reviewed, Refereed, Scholarly, Multidisciplinary and Open Access Journal | High Impact Factor 8.771 (Calculated by Google Scholar and Semantic Scholar | AI-Powered Research Tool | Indexing in all Major Database & Metadata, Citation Generator | Digital Object Identifier (DOI) |
| TITLE | Smart Fitness Coaching Platform for Real-Time Exercise Analysis |
|---|---|
| ABSTRACT | This study presents a real-time method for recognizing exercises using a Bidirectional Long Short-Term Memory (BiLSTM) model, specifically designed to work effectively in real-life conditions. Many existing exercise recognition systems depend heavily on synthetic datasets and only use raw joint coordinates. However, these coordinates can vary significantly depending on factors such as different users, camera positions, and environmental conditions. In addition, such systems often do not fully capture the temporal aspect of motion, which leads to reduced performance when used outside controlled environments. To address these issues, the proposed method combines joint angle features with raw spatial coordinates (x, y, z). This hybrid approach allows the model to better handle variations in viewpoint, body proportions, and positioning, while still maintaining precise spatial details of movement. The BiLSTM model is trained using sequences of 30 frames, enabling it to learn how movements change over time. To improve robustness, a diverse dataset was created by combining synthetic data from the InfiniteRep dataset with real-world videos from the Kaggle Workout/Exercises dataset and other online sources. The dataset includes four common exercises: squats, push-ups, shoulder presses, and bicep curls. During testing, the model achieved an accuracy of over 99% and showed strong performance when evaluated on separate real-world datasets, indicating good generalization ability. Comparisons with existing methods demonstrate clear improvements. In addition to exercise classification, the system also includes features such as automatic repetition counting, real-time posture correction, personalized workout and diet planning, and a chatbot that supports both physical and mental well-being. All these features are integrated into a single web-based application, providing a complete and user-friendly fitness solution. |
| AUTHOR | K. INDUMATHI, P. SATHISH KUMAR, A. SUDHAKAR, N. RAMESH PG Student, Department of Computer Science and Engineering, Bharathidasan Engineering College, Vellore, Tamil Nadu, India. Assistant Professor, Department of Computer Science and Engineering, Bharathidasan Engineering College, Vellore, Tamil Nadu, India. Head of the Department, Department of Computer Science and Engineering, Bharathidasan Engineering College, Vellore, Tamil Nadu, India. Assistant Professor, Department of Computer Science and Engineering, Bharathidasan Engineering College, Vellore, Tamil Nadu, India. |
| VOLUME | 184 |
| DOI | DOI: 10.15680/IJIRCCE.2026.1405087 |
| pdf/87_Smart Fitness Coaching Platform for Real-Time Exercise Analysis.pdf | |
| KEYWORDS | |
| References | [1] Maminiaina Alphonse Rafidison et al., “Image Classification Based on Light Convolutional Neural Network Using Pulse-Coupled Neural Network,” Computational Intelligence and Neuroscience, vol. 2023, pp. 1-17, 2023. [2] TusharRangari, Sudhanshu Kumar, Partha Pratim Roy, Debi Prosad Dogra, and Byung-Gyu Kim, “Video Based Exercise Recognition and Correct Pose Detection,” Multimedia Tools and Applications, vol. 81, no. 21, pp. 30267-30282, 2022. [3] Sushma V. et al., “Fitness Trainer Application Using Artificial Intelligence,” InternationalResearchJournalofEngineeringandTechnology(IRJET),vol.10,no.5, p.826, May 2023. [4] Grandel Dsouza, Deepak Maurya, and Anoop Patel, “SmartGymTrainerUsingHuman Pose Estimation,” 2022 IEEE International Conference for Innovation in Technology (INOCON), pp. 1-4, 2022. [5] Colin Arrowsmith, David Burns, Thomas Mak, Michael Hardisty, and Cari Whyne, “Physiotherapy Exercise Classification with Single-Camera Pose Detection and Machine Learning,” Sensors, vol. 23, no. 1, p. 363, 2022. [6] Boudjemaa Boudaa, Imen Bestani, and Noureddine Benadjrouda, “Graph Convolutional Networks for Designing Collaborative Filtering-Based Health Recommender Systems,” 2022 2nd International Conference on New Technologies of Information and Communication (NTIC), pp. 1-6, 2022. [7] Fotos Frangoudes, Maria Matsangidou, Eirini C. Schiza, Kleanthis Neokleous, and Constantinos S. Pattichis, “Assessing Human Motion During Exercise Using Machine Learning: A Literature Review,” IEEE Access, vol. 10, pp. 86874-86903, 2022. [8] Bang, G. S., & Park, S. B., “Workout Classification Using a Convolutional Neural Network in Ensemble Learning,” Sensors, vol. 24, no. 10, p. 3133, 2024. [9] Moran, A., Gebka, B., Golodshteyn, J., Beyer, A., Johnson, N., & Neuwirth, A., “MuscleVision:Real-TimeKeypointBasedPoseClassificationofPhysicalExercises,” vol. 1, no. 1, pp. 1-5, 2022 [10] Jindřich Adolf,PeterKán,TiareFeuchtner,BarboraAdolfová,Jaromír Doležal,& Lenka Lhotská. “OffiStretch: camera-based real-time feedback for daily stretching exercises.” The Visual Computer, vol. 41, no. 2 (February 2025), pp. 1555–1571. |
