Yoga Pose Classification Using ResNet-50 and Random Forest: A Hybrid Deep Learning Approach

The research paper presents a hybrid deep learning (DL) approach for yoga pose classification using the ResNet-50 model for feature extraction and a Random Forest classifier for final pose categorization. The dataset comprises 1,259 labeled images of 16 distinct yoga poses, such as chair pose, dolphin plank pose, and tree pose. Images were preprocessed by resizing to 224x224 pixels, followed by data augmentation techniques like random rotations and flips to improve generalization. ResNet-50, pre-trained on the ImageNet dataset, was fine-tuned for yoga pose recognition, producing high-dimensional feature vectors. These vectors were then used to train the Random Forest classifier, which provided accurate pose classification. The model was evaluated on 80 % training and 20 % validation data, achieving a validation accuracy of 96.06 %, with precision, recall, and F1-scores around 92 %. The combination of deep feature extraction and robust classification demonstrates strong performance in identifying yoga poses, despite challenges posed by class imbalance and subtle pose differences. The results highlight the model's ability to generalize across variations in pose execution and camera angles. The proposed ResNet-50 with Random Forest model offers a scalable and effective solution for automated yoga pose recognition.