A Highlighted Source of Big Data for Artificial Intelligence: The Growing Impact of Automated Tracking Systems
As seen in Chapters 1 and 3, machine learning has become increasingly popular in computer vision-based applications due to the state-of-the-art results reachable in the image classification, object detection, and natural language processing domains of performance (Elhoseny, 2020). Video watching applications are becoming increasingly important for analysing individual performance, and video object tracking is an essential task because it generates interrelated temporal data about the moving objects in a film clip (Luo et al., 2018). Over the last few years, systems capable of tracking the movement of players and the ball in different sports (e.g., football, rugby, handball) have multiplied, which provides a detailed characterization and analysis of the behaviour of athletes and teams.
The usefulness of this type of technology was well demonstrated by Chambers et al. (2019), who developed and validated a scrum algorithm to automatically detect scrum events in rugby. Such innovation allows a significant efficiency in the process of players’ performance analysis, since previously, this type of analysis was only possible through the use of notational analysis techniques which proved to be very time consuming. Using automated tracking system technology and analysis allows coaches and sport scientists to determine the physical load associated with these contact events, which should improve player-training process and reduce the risk of injury.
In volleyball, Hsu et al. (2016) developed an automatic system capable of extracting play-by-play periods and locating players using a novel 2D histogram-based approach, which proved to be effective not only for obtaining tactical information but also for collecting descriptive performance statistics during competition. Additionally, Gomez et al. (2014) compared two player tracking methods (a classical particle filter and a rigid grid integral histogram tracker) in beach volley. They concluded that contrary to the previous literature where tracking results of over 90% were verified, in this study tracking accuracy of the ball was 54.2% for the trajectory growth method and 42.1% for the Hough line detection method. The difficulties in accurately detecting the ball trajectories had also been previously reported by Chen et al. (2012) in baseball, and by Leo et al. (2013) in soccer. More recently, Takahashi et al. (2018) reported to have a successful experience in improving a system to automatically detect the movement of the ball in soccer, integrating tracking results from multi-view cameras in real time.
Over the last decades there has been an exponential increase in soccer video analysis research (Leo et al., 2013). Using Bayesian methods, some investigators (Martinez-del-Rincon et al., 2009; Motoi et al., 2012) tried to develop some game tracking systems that have revealed some functionality in detecting in-game events. These models seem to be important not only for coaches and practitioners but also for an increasing number of fans at home or in the stadium who may be interested in consuming more ‘live’ detailed, performance information.