Enhanced image technique for Yoga Pose Recognition based on MediaPipe and YOLOv8

Movement is the foundation of daily physical activities. Physical movement also includes exercise for health. Yoga is another method that people are more interested in nowadays, especially after the COVID-19 outbreak. Because it does not require much space, it can be exercised in the room without traveling. Although yoga is a slow exercise, yoga practitioners may still be affected and injured by incorrect body movements. Doing yoga in stretching or twisting poses that are different from standard yoga poses. Research on studying correct yoga pose recognition has been popular in recent years. Therefore, researchers propose enhance image technique to improve the performance of yoga pose recognition by comparing the MediaPipe and YOLOv8-Pose models through deep learning. YOLOv8 has many techniques to improve the performance of point detection, such as adjusting the image contrast with the Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm and improving the image color quality with the Multi-Scale Retinex with Color Restoration (MSRCR) algorithm. This research fuses the two algorithms to improve the efficiency of the MediaPipe and YOLOv8-Pose datasets. They have different numbers of keypoints, but they can accurately detect poses and identify keypoints at body joints. Our comparative analysis on yoga pose identification shows that MediaPipe has the highest accuracy in identifying yoga poses from body joint detection, with a mean accuracy (mAP) of 94.9% and YOLOv8-Pose of 92.0%, indicating that the MediaPipe model is more effective in identifying yoga poses in this research.