Human Exercise Posture Detection Using MediaPipe and Machine Learning

Human posture estimation is a type of computer vision technology that recognizes and estimates the positions of human joints in images and videos. Various studies have been conducted on various algorithms, technologies, and methodologies, ranging from geodesic distant feature to hierarchical context network, and state of the art method, to the most recent and advanced deep learning techniques such as convolutional neural networks, Hourglass network, multi-stage approach, and so on. In this study paper, we will discuss various strategies and compare them to one another. Apart from this, we are going to use Mediapipe framework to gather landmarks of human pose and based on those landmarks we are going to extract various features like joint angles, 3d distance between various key points of human posture, etc. Based on these features we will be using a machine learning model to train a large dataset of these features and the corresponding pose. The model will be able to detect the human exercise postures like jumping jacks, pull-ups, push-ups, squats and sit-ups. The machine learning model is trained using decision trees, gradient boosting algorithm, and naíve bayes algorithm which will enhance the precision of the predicted pose.