Date on Master's Thesis/Doctoral Dissertation
Computer Engineering and Computer Science
Computer Science, MS
Committee Co-Chair (if applicable)
computer vision; machine learning; baseball
Lokator is a baseball training system designed to document pitch location while teaching pitch command, selection and sequencing. It is composed of a pitching target and a smartphone app. The target is divided into a set of zones to identify the pitch location. The main limitation of the current system is its reliance on the user's feedback. After each throw, the pitcher or the coach needs to identify and report the target's zone that was hit by the ball by just relying on the naked eye. The purpose of this thesis is to investigate the possibility of using computer vision technology to automate the pitch analysis in baseball and improve the usability and accuracy of the Lokator system. Towards this goal, we have developed, implemented and tested a computer vision-based software system that adds the following contributions to the Lokator system: Automated and accurate reading of the pitch location on the target. Automated and accurate estimation of the velocity of the ball. Provide contextual information about the pitch, such as vertical movement of the ball which can indicate late breaking. Replace the target by a catcher and estimate the pitch location using a virtual target. We have tested the software on a large set of recording. Those recordings are from indoor and outdoor environments with various illumination conditions and different backgrounds. The software was also tested on videos with softball pitches. To estimate the accuracy of the software, the sponsor gave us a set of 15 videos that include a total of 144 pitches along with the hit location of each pitch. Another set of 8 videos were provided to measure the accuracy of our software in terms of speed calculation.
Moalla, Mahdi, "Developing computer vision technology to automate pitch analysis in baseball." (2017). Electronic Theses and Dissertations. Paper 2686.