Smart Yoga Trainer: Real-Time Pose Recognition and Feedback Using Similarity Metrics

The COVID-19 pandemic accelerated the shift to online platforms worldwide for education, training, shopping, and remote work. With the increasing prominence of artificial intelligence and home-based yoga practices, an opportunity now arises for the translation of traditional in-person yoga teaching to accessible virtual training. In this paper, the authors introduce an AI-driven system for detecting and correcting yoga poses in real time using MediaPipe. It extracts skeletal key points from live camera footage and evaluates the accuracy of posture by analyzing joint angles. Additionally, YOLOv5 is used to further improve the detection of the alignment of the body by utilizing the YOLO Yoga Dataset.v1i.yolov5 with 1,013 images, where augmentation was applied to increase the robustness of the images. The integration of deep learning with real-time pose estimation renders the system a virtual yoga assistant that can give instantaneous feedback on alignment and encourage home-based practice in safer ways.