PoseNet based Model for Estimation of Karate Poses

Abstract

In the domain of computer vision, human pose estimation is becoming increasingly significant. It's one of the most compelling areas of research, and it's gaining a lot of interest due to its usefulness and flexibility in a variety of fields, including healthcare, gaming, augmented reality, virtual trainings and sports. Human pose estimation has opened a door of opportunities. This paper proposes a model for estimation and classification of karate poses which can be used in virtual karate posture correction and trainings. A pretrained model, PoseNet has been used for pose estimation using the results of which the angles between specific joints are calculated and fed into a K-Nearest Neighbors Classifier to classify the poses. The results obtained show that the model achieves an accuracy of 98.75%.