YogiCombineDeep: Enhanced Yogic Posture Classification Using Combined Deep Fusion of VGG16 and VGG19 Features

Abstract

Yoga garnered significant attention during this pandemic based on its physical benefits in recent years. Regular yoga practice can enhance both physical and mental health. However, some body parts are occluded due to the significant variations in specific asanas with complicated posture formations and their backgrounds, making yogic posture detection more complex. This study establishes the classification of yoga postures in the yoga-16 dataset utilizing the combination of deep learning and machine learning approaches. A pre-trained CNN architecture VGG16 and VGG19 collects the deep features from the images of yoga postures. Then, the collected features are combined and entered into classifiers to train and assess the outcome of yoga posture classification. To classify the yogic postures from the collected yoga-16 dataset, the proposed model includes logistic regression, support vector machines, random forest, and extra tree classifiers. The proposed approach examined the yoga-16 dataset containing 16 classes and 6561 images. The proposed combined deep-fused approach utilizing Linear SVM yields better results than all the existing yogic posture classification models with outstanding scores of 99.94% precision, 99.94% recall, 100% f1-score, and 99.92% accuracy, respectively. The results show that the proposed approach is effective at attaining excellent performance in yogic posture classification. Performance comparisons with the most recent models have also been listed.

Keywords

• Yoga pose classification
• VGG16
• VGG19
• machine learning
• deep learning models
• combined features