Page 37 - 2024F

30 UEC Int’l Mini-Conference No.53

Baseball Pitch Classification based

on Skeleton Data

2
3
Sergio E. HUESCA-FLORES 1,2 , Mariko NAKANO , Gibran BENITEZ-GARCIA , and Hiroki
TAKAHASHI 3
1 UEC Exchange Study Program (JUSST Program)
2 National Polytechnic Institute, Mexico City, Mexico
3 Informatics Department
The University of Electro-Communications, Tokyo, Japan

Abstract
In baseball, accurate pitch type classification is essential for strategic decision-making by teams and
analysts. Traditional methods rely heavily on ball trajectory tracking using radar and high-speed cameras,
which are costly and accessible only in professional leagues. In contrast, this paper presents a skeleton-
based approach to classify pitch types using pose estimation and spatial-temporal modeling. We extract key
joint coordinates of the pitcher using OpenPose and model their body movements over time using a Spatial-
Temporal Graph Convolutional Network (ST-GCN). Our method is evaluated on the publicly available MLB-
YouTube dataset, achieving 68.2% accuracy in classifying six pitch types, and outperforming state-of-the-art
methods that rely on full-frame data with 3D CNNs. By focusing exclusively on pitcher’s skeletal information
through graph-based modeling, our approach improves classification performance, while showing robustness
in binary tasks, reaching 85.7% accuracy for fastball detection and 80.8% distinguishing fast versus slow
pitches. Our method’s performance underscores the effectiveness of relying on body mechanics for pitch
classification, demonstrating that accurate results can be achieved without the need for costly ball trajectory
data.
Keywords: Pitch Classification, Pose Estimation, ST-GCN, Baseball Analytics, Action Recognition

I. INTRODUCTION often aim to deceive batters by making small ad-
Statistics in baseball are vital for teams, an- justments in their body mechanics. These minimal
alysts, and fans, shaping strategic decisions and variations make it difficult to distinguish between
player evaluations. Classifying pitch types is par- pitches like fastballs, changeups, and sliders, even
ticularly important as it helps teams analyze a for experienced analysts [2], [3]. Recent research,
pitcher’s repertoire and allows hitters to prepare such as that by Gomaz et al. [4], has shown
for matchups. Traditionally, this classification has that body mechanics, specifically adjustments in
relied on technologies like the Statcast system [1], posture, arm angles, and joint timing, are crucial for
which utilizes a combination of radar and high- differentiating between pitches. Shifting the focus
speed cameras to track both the ball flight and from tracking the ball to analyzing the pitcher’s
the pitcher’s arm movements. Additionally, manual body movements offers a promising alternative, by
annotations are often employed to further enhance concentrating on the biomechanics of the pitcher
data collection. While these methods provide de- rather than solely on the ball’s trajectory, deeper
tailed and accurate information, they are costly, insights can be gained into the distinguishing fea-
require significant maintenance, and are typically tures of each pitch type. Therefore, in this paper,
limited to professional settings like the Major we propose to isolate the pitcher’s pose to analyze
League Baseball (MLB). their body mechanics as a skeleton graph. To do
The challenge of pitch classification lies in the this, we first apply multi-person pose estimation
subtle differences between pitch types, as pitchers using OpenPose [5] to extract joint coordinates
for all individuals detected in each frame. From
The author is supported by JASSO Scholarship. these poses, we isolate the pitcher’s pose, and