Automatically recognizing strategic cooperative behaviors in various situations of a team sport

Abstract

Understanding multi-agent cooperative behavior is challenging in various scientific and engineering domains. In some cases, such as team sports, many cooperative behaviors can be visually categorized and labeled manually by experts. However, these actions which are manually categorized with the same label based on its function have low spatiotemporal similarity. In other words, it is difficult to find similar and different structures of the motions with the same and different labels, respectively. Here, we propose an automatic recognition system for strategic cooperative plays, which are the minimal, basic, and diverse plays in a ball game. Using player’s moving distance, geometric information, and distances among players, the proposed method accurately discriminated not only the cooperative plays in a primary area, i.e., near the ball, but also those distant from a primary area. We also propose a method to classify more detailed types of cooperative plays in various situations. The proposed framework, which sheds light on inconspicuous players to play important roles, could have a potential to detect well-defined and labeled cooperative behaviors.

Physically-interpretable classification of biological network dynamics for complex collective motions

“…For hand-crafted features, we compute simple features to demonstrate the validity of our methods in this paper. The use of a more customised framework [44] which includes the detection of a few related people might improve specific classification performance. However, in this paper, we proposed a more generalised classification framework for global dynamics of collective motions, which successfully performed two different recognition tasks (team-defence and offence) by selecting only the elements of the Graph DMD modes as feature vectors.…”

“…Prior to data segmentation, we used a custom-made automatic individual play-detection system to detect shots using positional data similar to that used in previous studies [22,38,44]. We analysed an attack-segment defined as the period that begins when all players enter the attacking side of the court and ends before a shot (we analysed only the attack-segment finishing with a shot).…”

“…Screen-play is the basic and minimal strategic cooperative play in basketball [44], in which an attacker stands on the course of defence player like a 'screen' and prevents the defender from defending another attacker in a legal way. We labelled attack-segments into an offence with (at least including a screen-play) and without screen-play.…”

“…Note that we focused on the global network dynamics of collective motions, thus we did not segment and label screen-plays themselves (i.e. in a spatio-temporally accurate sense such as in the previous study [44]).…”

Dynamic mode decomposition in vector-valued reproducing kernel Hilbert spaces for extracting dynamical structure among observables

“…Next, we evaluated our method using a example with unknown global dynamics, because in some real-world (especially biological) data, the true global spatiotemporal structure is sometimes unknown [35,53]. For evaluation, here we used well-known collective motion models [54] with simple local rules to generate multiple distinct group behavioral patterns (Figure 3a): swarm, torus, and parallel behavioral shapes.…”