April 26th, 2024 – 11:30am EC4-2101A

Presenter: Jerrin Bright

In this seminar, I will present my work on PitcherNet, a system that analyze baseball pitchers by explicitly modelling the pitcher directly from live broadcasts, offering real-time insights into a pitcher’s performance without the need for special equipment or pre-recorded data. By analyzing broadcast video, PitcherNet learns the pitcher kinematics, and extracts key pitching metrics including set position, pitch velocity, release point and pitch extension. This allows for immediate insights into a pitcher’s mechanics, allowing for real-time adjustments and improved performance. The talk will focus on the challenges of using broadcast footage, including motion blur and self-occlusion, and how PitcherNet overcomes these hurdles to generate accurate and reliable 3D human models.