Hybrid Convolutional Neural Network-Long Short Term Memory based Real-Time Physical Training Monitoring System using Browser/Server Architecture

The growing need for efficient fitness management has given rise to intelligent systems that can track and analyse physical training activities in real time. This paper introduces B/S Mode (Browser/Server architecture), combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks to enhance the accuracy and timeliness of motion recognition. Conventional systems are greatly dependent on manual monitoring or simple sensor data analysis, tending to lead to low scalability, low real-time feedback, and low adaptability to intricate movement patterns. Current methods mostly employ either CNN for spatial feature extraction or LSTM for temporal sequence modeling, but rarely both simultaneously. Our hybrid CNN-LSTM model efficiently processes spatiotemporal training data, identifying dynamic changes in physical activities. Relative to conventional SVM and isolated CNN models, our system has an accuracy of 99.4% and F1-score of 99.8%, indicating large performance and flexibility improvements. Cross-platform availability and light client interaction are guaranteed through the B/S mode. The system's enhanced performance indicates its potential for real-time athlete tracking, customized feedback, and compatibility with smart fitness platforms.