Baseball pitch type recognition based on broadcast videos

Reed Chen; Dylan Siegler; Michael Fasko; Shunkun Yang; Xiong Luo; Wenbing Zhao

doi:10.1007/978-981-15-1925-3_24

Baseball pitch type recognition based on broadcast videos

Reed Chen
, Dylan Siegler
, Michael Fasko
, Shunkun Yang
, Xiong Luo
, Wenbing Zhao

Electrical and Computer Engineering

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

7 Scopus citations

Abstract

In this paper, we report our work on baseball pitch type recognition based on broadcast videos using two-stream inflated 3D convolutional neural network (I3D). To improve the state-of-the-art of research, we developed our own high-quality dataset, trained and tuned the I3D model extensively, primarily combating the problem of overfitting while still trying to improve final validation accuracy. In the end, we are able to achieve an accuracy of 53.43% ± 3.04% when oversampling and 57.10% ± 2.99% when not oversampling, which is a significant improvement over the published best result of an accuracy of 36.4% on the same six pitch type classes.

Original language	English
Title of host publication	Communications in Computer and Information Science
Editors	Huansheng Ning
Place of Publication	che
Publisher	Springer
Pages	328-344
Number of pages	17
Volume	1138 CCIS
ISBN (Print)	9789811519246
DOIs	https://doi.org/10.1007/978-981-15-1925-3_24
State	Published - Jan 1 2019
Event	3rd International Conference on Cyberspace Data and Intelligence, Cyber DI 2019, and the International Conference on Cyber-Living, Cyber-Syndrome, and Cyber-Health, CyberLife 2019 - Beijing, China Duration: Dec 16 2019 → Dec 18 2019

Publication series

Name	Communications in Computer and Information Science
Publisher	Springer
Volume	1138 CCIS
ISSN (Print)	18650929
ISSN (Electronic)	18650937

Conference

Conference	3rd International Conference on Cyberspace Data and Intelligence, Cyber DI 2019, and the International Conference on Cyber-Living, Cyber-Syndrome, and Cyber-Health, CyberLife 2019
Country/Territory	China
City	Beijing
Period	12/16/19 → 12/18/19

Keywords

Baseball pitch type recognition
Overfitting
Regularization
Support vector machine
Two-stream inflated 3D convolutional neural network

Access to Document

10.1007/978-981-15-1925-3_24

Cite this

@inproceedings{98c8fe860c8f4fcbabaeb5ecf5caed1d,

title = "Baseball pitch type recognition based on broadcast videos",

abstract = "In this paper, we report our work on baseball pitch type recognition based on broadcast videos using two-stream inflated 3D convolutional neural network (I3D). To improve the state-of-the-art of research, we developed our own high-quality dataset, trained and tuned the I3D model extensively, primarily combating the problem of overfitting while still trying to improve final validation accuracy. In the end, we are able to achieve an accuracy of 53.43\% ± 3.04\% when oversampling and 57.10\% ± 2.99\% when not oversampling, which is a significant improvement over the published best result of an accuracy of 36.4\% on the same six pitch type classes.",

keywords = "Baseball pitch type recognition, Overfitting, Regularization, Support vector machine, Two-stream inflated 3D convolutional neural network",

author = "Reed Chen and Dylan Siegler and Michael Fasko and Shunkun Yang and Xiong Luo and Wenbing Zhao",

year = "2019",

month = jan,

day = "1",

doi = "10.1007/978-981-15-1925-3\_24",

language = "English",

isbn = "9789811519246",

volume = "1138 CCIS",

series = "Communications in Computer and Information Science",

publisher = "Springer",

pages = "328--344",

editor = "Huansheng Ning",

booktitle = "Communications in Computer and Information Science",

note = "3rd International Conference on Cyberspace Data and Intelligence, Cyber DI 2019, and the International Conference on Cyber-Living, Cyber-Syndrome, and Cyber-Health, CyberLife 2019 ; Conference date: 16-12-2019 Through 18-12-2019",

}

Chen, R, Siegler, D, Fasko, M, Yang, S, Luo, X & Zhao, W 2019, Baseball pitch type recognition based on broadcast videos. in H Ning (ed.), Communications in Computer and Information Science. vol. 1138 CCIS, Communications in Computer and Information Science, vol. 1138 CCIS, Springer, che, pp. 328-344, 3rd International Conference on Cyberspace Data and Intelligence, Cyber DI 2019, and the International Conference on Cyber-Living, Cyber-Syndrome, and Cyber-Health, CyberLife 2019, Beijing, China, 12/16/19. https://doi.org/10.1007/978-981-15-1925-3_24

Baseball pitch type recognition based on broadcast videos. / Chen, Reed; Siegler, Dylan; Fasko, Michael et al.
Communications in Computer and Information Science. ed. / Huansheng Ning. Vol. 1138 CCIS che: Springer, 2019. p. 328-344 (Communications in Computer and Information Science; Vol. 1138 CCIS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Baseball pitch type recognition based on broadcast videos

AU - Chen, Reed

AU - Siegler, Dylan

AU - Fasko, Michael

AU - Yang, Shunkun

AU - Luo, Xiong

AU - Zhao, Wenbing

PY - 2019/1/1

Y1 - 2019/1/1

N2 - In this paper, we report our work on baseball pitch type recognition based on broadcast videos using two-stream inflated 3D convolutional neural network (I3D). To improve the state-of-the-art of research, we developed our own high-quality dataset, trained and tuned the I3D model extensively, primarily combating the problem of overfitting while still trying to improve final validation accuracy. In the end, we are able to achieve an accuracy of 53.43% ± 3.04% when oversampling and 57.10% ± 2.99% when not oversampling, which is a significant improvement over the published best result of an accuracy of 36.4% on the same six pitch type classes.

AB - In this paper, we report our work on baseball pitch type recognition based on broadcast videos using two-stream inflated 3D convolutional neural network (I3D). To improve the state-of-the-art of research, we developed our own high-quality dataset, trained and tuned the I3D model extensively, primarily combating the problem of overfitting while still trying to improve final validation accuracy. In the end, we are able to achieve an accuracy of 53.43% ± 3.04% when oversampling and 57.10% ± 2.99% when not oversampling, which is a significant improvement over the published best result of an accuracy of 36.4% on the same six pitch type classes.

KW - Baseball pitch type recognition

KW - Overfitting

KW - Regularization

KW - Support vector machine

KW - Two-stream inflated 3D convolutional neural network

UR - https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85076907705&origin=inward

UR - https://www.scopus.com/inward/citedby.uri?partnerID=HzOxMe3b&scp=85076907705&origin=inward

U2 - 10.1007/978-981-15-1925-3_24

DO - 10.1007/978-981-15-1925-3_24

M3 - Conference contribution

SN - 9789811519246

VL - 1138 CCIS

T3 - Communications in Computer and Information Science

SP - 328

EP - 344

BT - Communications in Computer and Information Science

A2 - Ning, Huansheng

PB - Springer

CY - che

T2 - 3rd International Conference on Cyberspace Data and Intelligence, Cyber DI 2019, and the International Conference on Cyber-Living, Cyber-Syndrome, and Cyber-Health, CyberLife 2019

Y2 - 16 December 2019 through 18 December 2019

ER -

Baseball pitch type recognition based on broadcast videos

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Cite this