Interpretable Pre-Release Baseball Pitch Type Anticipation from Broadcast 3D Kinematics

2026-03-05

AI Summary

This study addresses the challenge of predicting eight distinct types of baseball pitches solely from a pitcher’s pre-release body movements, without relying on ball trajectory data. Leveraging a large-scale dataset of 119,561 professional pitches represented as monocular 3D pose sequences, the authors propose an end-to-end pipeline integrating diffusion model–driven pose estimation, automated pitch event detection, biomechanical feature extraction, and gradient-boosted classification over a 229-dimensional kinematic feature space. The approach achieves 80.4% accuracy on real-world data and, for the first time, quantifies the relative contributions of upper and lower body motion—revealing that the upper body accounts for 64.9% of predictive power. Key biomechanical insights include the critical roles of wrist position and trunk tilt, while also demonstrating that fastball variants defined by grip cannot be reliably distinguished by kinematics alone.

Technology Category

• Computer Vision: Biometrics, Face, Gesture & Pose
• Machine Learning: Evaluation and Analysis
• Intelligent Robots: Manipulation

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Web Mining and Content Analysis: Large pretrained models with web data
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

How much can a pitcher's body reveal about the upcoming pitch? We study this question at scale by classifying eight pitch types from monocular 3D pose sequences, without access to ball-flight data. Our pipeline chains a diffusion-based 3D pose backbone with automatic pitching-event detection, groundtruth-validated biomechanical feature extraction, and gradient-boosted classification over 229 kinematic features. Evaluated on 119,561 professional pitches, the largest such benchmark to date, we achieve 80.4\% accuracy using body kinematics alone. A systematic importance analysis reveals that upper-body mechanics contribute 64.9\% of the predictive signal versus 35.1\% for the lower body, with wrist position (14.8\%) and trunk lateral tilt emerging as the most informative joint group and biomechanical feature, respectively. We further show that grip-defined variants (four-seam vs.\ two-seam fastball) are not separable from pose, establishing an empirical ceiling near 80\% and delineating where kinematic information ends and ball-flight information begins.

Problem

• pitch type anticipation
• 3D kinematics
• baseball pitching
• interpretable prediction
• biomechanics

Innovation

• 3D pose estimation
• pitch type anticipation
• biomechanical features
• diffusion-based modeling
• interpretable AI