Bayesian event detection for sport games with hidden Markov model

Abstract

Event detection can be defined as the problem of detecting when a target event has occurred, from a given data sequence. Such an event detection problem can be found in many fields in science and engineering, such as signal processing, pattern recognition, and image processing. In recent years, many data sequences used in these fields, especially in video data analysis, tend to be high dimensional. In this paper, we propose a novel event detection method for high-dimensional data sequences in soccer video analysis. The proposed method assumes a Bayesian hidden Markov model with hyperparameter learning in addition to the parameter leaning. This is in an attempt to reduce undesired influences from ineffective components within the high-dimensional data. Implemention is performed by Markov Chain Monte Carlo. The proposed method was tested against an event detection problem with sequences of 40-dimensional feature values extracted from real professional soccer games. The algorithm appears functional.

Toward Automatically Labeling Situations in Soccer

“…There are several recent studies aiming to detect basic events directly out of video footage (Ekin et al, 2003 ; Wickramaratna et al, 2005 ; Kolekar and Palaniappan, 2009 ) or positional data (Zheng and Kudenko, 2010 ; Motoi et al, 2012 ; Richly et al, 2016 ; Stein et al, 2019 ) and others focus on the identification of sophisticated tactical patterns (Hobbs et al, 2018 ; Andrienko et al, 2019 ; Shaw and Sudarshan, 2020 ; Anzer et al, 2021 ; Bauer and Anzer, 2021 ). The proposed approaches provide useful solutions for their respective tasks.…”

Event Detection and Recognition Using HMM with Whistle Sounds

“…On the other hand, Motoi [6] detected the sports event using Hidden Markov Model. This method employed General Hidden Markov Model, which could express the time dependency in time series data.…”

Modeling state duration and emission dependence in hidden Markov and hidden semi-Markov models

“…HMMs and HSMMs are an appropriate approach for analyzing these high temporal resolution data when the observations are considered to be related to a latent or unobserved process. Among the first areas to take advantage of these models was speech recognition, but they have since been applied in other areas such as activity recognition (Duong et al, 2005;Natarajan and Nevatia, 2007), event recognition in videos (Motoi et al, 2012;Xie et al, 2004), animal movement (Scharf et al, 2016;Leos-Barajas et al, 2017), and animal behavior (Ruiz-Suarez et al, 2022). For a more comprehensive list of applications of these models see Mor et al (2021), and Chapter 9 in Yu (2016).…”

“…These models consider a Markov chain of unobserved states and a sequence of observations dependent on the states (Yu, 2010). Motoi et al (2012) used a Bayesian HMM to create metadata for sports games through event detection, and their method was evaluated using video data of soccer games to detect events such as kick offs, corner kicks, or goal kicks. Thomas et al ( 2010) performed activity recognition using both an HMM and a semi-Markov model (SMM) applied to swimming data collected with a wearable sensor, where the segmentation of the session improved evaluation of the training.…”