AI Personal Trainer for Lateral Raises and Shoulder Presses Exercises

Resistance training has picked up popularity due to its various benefits, including improvements in muscle quality, control, balance, and coordination. In any case, this sort of training also postures certain risks, such as unexpected injuries to the back, muscles, ligaments, and tendons. This paper introduces an AI personal trainer that utilizes a webcam or mobile camera to guide users in performing shoulder exercises, such as lateral raises and shoulder presses, correctly. The AI trainer tracks repetitions, classifies exercises, calculates exercise speed, and displays joint coordinates graph using the $1 Unistroke Recognizer to demonstrate the user's range of motion after each session, allowing for comparison with a professional trainer’s graph. The AI trainer uses Mediapipe to extract 33 pose body coordinates and saves them to CSV files, resulting in a dataset of 81,937 samples. Two additional CSV files were generated for each exercise, resulting in a dataset of 16,496 samples, categorizing landmarks as 'up' or 'down' labels to detect hand movements and assist with repetition counting. The application offers real-time feedback and notifies the user when to slow down to avoid injury. Experimental results show that the KNN model achieved 100% accuracy for Exercise Type CSV, and the Random Forest Classifier achieved the highest accuracy of 99.80% and 98.49% for Lateral Raises and Shoulder Press Up and Down CSV datasets respectively. Consequently, these models were selected for implementation due to their high reliability in exercise classification.