Calisthenics Movement Recognition System Using Reinforcement Model Based on Multi Task Learning

In recent days, the classification of calisthenics movements is essential for improving exercise efficacy, injury prevention, and personalized training. Calisthenics is a dynamic activity which combines both rhythmic movements along with cardiovascular benefits. Previous researchers have suggested various traditional methods that depend on manual observation and thus fail to capture significant motion patterns. Therefore, this paper suggests incorporating Multi Task Learning (MTL) model to classify the calisthenics movements effectively. Initially, the data is inherited from University of Idaho Physical Rehabilitation Movement Dataset (UI-PRMD) which contains different calisthenics movement and poses information, and then the collected data is processed through effectively techniques such as data augmentation and Contrast Limited Adaptive Histogram Equalization (CLAHE). Further, the processed data is fed into Saliency Based Selection (SBS) model for extraction of optimal features. Finally, the extracted features are embedded into the MTL model for effective classification of calisthenics movement. The proposed MTL model acquired higher results with the accuracy of 99.68%, Positive Predictive Value (PPV) of 99.14%, Sensitivity of 98.81 % and F1-score of 98.97% when compared with the existing hybrid model integrated with Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM).