MARKERLESS TRACKING OF THE BIOMECHANICS OF A PITCHER’S ARM USING DEEP LEARNING

Author

Ryan Netemeyer, Southern Illinois University CarbondaleFollow

Date of Award

12-1-2019

Degree Name

Master of Science

Department

Mechanical Engineering

First Advisor

Nilufar, Sabrina

Abstract

The objective of this research is to use machine learning to accurately track and label body parts of a pitcher’s arm during the throwing sequence. By tracking the labeled body parts, players and coaches can decrease injuries and increase performance. The goal is to bring this technology to the college baseball level. Ten different videos of five pitches were used in the research. Five videos came from a camera that showed the armside view of the pitcher and five videos came from a camera that showed the backside view of the pitcher. The body parts of the arm labeled are the wrist, forearm, elbow, bicep, and shoulder. By tracking and labeling these body parts the user can look for the patterns of the body parts from pitch to pitch for any inconsistencies in the pitcher’s arm. The machine learning was done through DeepLabCut. The user labeled the body parts for the machine learning to learn, track, and label the body parts to track and label. After labeling the machine learning tracked the body parts on each individual frame. The machine learning had an error of 1.15 pixels for the armside view videos and 1.45 pixels from the backside view videos. The body parts of each videos were graphed in relation to the x and y-positions of each frame. Each body part from each video were graphed to show how each body part moves in relation to each pitch. Players and coaches can look at the graphs of the labeled body parts to see any inconsistencies present.