Abstract:
Fitness exercises are highly beneficial for personal health, but they can be ineffective or even dangerous if performed improperly. One common reason for this is a lack of proper form or pose while exercising. While there are currently no practical solutions avail- able for evaluating and detecting mistakes in exercise form, there have been a numberof strategies developed for extracting human key points and classifying their motion in time series data. To address this issue, an automated workout guide called Workout-Vis has been developed. This application uses the latest techniques in pose estimation and machine learning to detect a user’s exercise pose and provide personalized feedback on its correctness. It allows users to work out in real-time and track their progress over time Proposed project uses a combination of machine learning, deep learning, data science, and data preprocessing. It utilizes state-of-the-art pose estimation to detect a user’s pose and evaluate the vector geometry of the pose during an exercise to provide useful feedback. A dataset of exercise videos showing both correct and incorrect form basedon personal training guidelines was recorded and fed into the pose estimator to extractkey points on the human body. Machine learning and deep learning models were then built to detect the exercise in the video feed using these key points and evaluate the performance of the workout activity. Workout-Vis focuses on multiple common exercises and provides valuable analysis of their execution in dynamic environments. Overall, the goal of Workout-Vis is to provide a helpful tool for improving the effectiveness and safetyof fitness exercises. By detecting and correcting mistakes in real- time, it aims to assist users in achieving their fitness goals more efficiently and effectively
