Analyzing Deadlift Form with Bio-mechanical Linkage Data

Abstract

This study explores an approach for analyzing deadlift forms using biomechanical linkage data and neural networks. Methods such as personal trainers and manual corrections can be costly and ineffective without the right tools, creating significant injury risks. By using Openpose pose estimation and feed-forward neural networks to classify deadlift form and deviations from proper form, we developed a system that has nearly 100% accuracy. Because these results are often hard to understand, a custom GPT was created to transform the data to be readable for people to take action and fix their form. The approach demonstrates the effectiveness of machine learning and pose estimation working together in strength training and proves how it can be used in many other applications of exercise.