No abstract

Image-based sports analytics enable automatic retrieval of key events in a game to speed up the analytics process for human experts. However, most existing methods focus on structured television broadcast video datasets with a straight and fixed camera having minimum variability in the capturing pose. In this paper, we study the case of event detection in sports videos for unstructured environments with arbitrary camera angles. The transition from structured to unstructured video analysis produces multiple challenges that we address in our paper. Specifically, we identify and solve two major problems: unsupervised identification of players in an unstructured setting and generalization of the trained models to pose variations due to arbitrary shooting angles. For the first problem, we propose a temporal feature aggregation algorithm using person re-identification features to obtain high player retrieval precision by boosting a weak heuristic scoring method. Additionally, we propose a data augmentation technique, based on multi-modal image translation model, to reduce bias in the appearance of training samples. Experimental evaluations show that our proposed method improves precision for player retrieval from 0.78 to 0.86 for obliquely angled videos. Additionally, we obtain an improvement in F1 score for rally detection in table tennis videos from 0.79 in case of global frame-level features to 0.89 using our proposed player-level features. Please see the supplementary video submission at https://ibm.biz/BdzeZA.

The purpose of this study was to examine the factors influencing the behavior of the baseball during and after the impact with a wooden bat. A finite element approach was used to model the behavior of a baseball and a wooden bat at the impact and the linear and angular velocities of the ball after the impact were determined for given sets of impact condition. The simulation outcome was validated with 22 sets of actual measurements taken during a "free batting practice" performed by the members of a collegiate baseball team.

In table tennis rallies between elite players, rapid swing movements must be achieved in a posture that is difficult to maintain while maintaining body agility. Players are required to have both a high racket speed and quick swing ability. On the other hand, a large movable range during a back swing may be disadvantageous for realizing a rapid, high-speed swing. Against this background, we sought to investigate the shoulder and torso joint movement characteristics of elite table tennis players when performing topspin forehands with 3 types of body movement, and to clarify the relationship between the maximum range of motion and swing skill. Ten male elite players participated in this experiment. Each performed 3 types of stroke trial (chance ball task, step over task, and step around task) and the active ranges of motion in the shoulder and torso joints (ROM trial) were measured. The three-dimensional coordinates of body landmarks (thorax, pelvis, and dominant humerus) and the racket during measurement were acquired using an optical motion capture system. The relationship between the maximum backward rotation angle at the shoulder and torso joint, the racket speed at ball impact, and the start time of the forward swing in each segment or joint were investigated. There was no significant difference in the maximum horizontal abduction angle at the shoulder joint between ROM trial and stroke trial (p > 0.2). Maximum torso backward rotation angle in ROM trial was significantly larger than that in the stroke trial (p < 0.05). In ROM trial and each stroke trial, the anatomical and functional flexibility of the shoulder and torso joints did not positively affect the racket speed and swing time. Rather, performance tended to be better during smaller backswings. Therefore, table tennis players require muscle strength or a technique that increases joint stiffness during racket swings.

Baseball pitchers use various pitch types to reduce hitting accuracy, but little is understood of the practical strategy of using visuomotor skills and timing control to respond to different pitches. This study examined 1) effectiveness of pitch type combinations, and 2) relationship between the presence and absence of advance information about the next pitch and the timing error. Twenty-six high school baseball players hit a ball launched from a pitching machine in a combination of fastballs (34.3±1.3 m�s -1 ), curveballs (25.4±1.0 m�s -1 ), and slowballs (25.5±0.9 m�s -1 ). Each participant performed three conditions. (1) Continuity condition (15 trials), in which the same pitch type was thrown five times consecutively.(2) Random condition (30 trials), in which pitch type was not preliminarily conveyed to the participants. (3) Open condition (20 trials), in which the next pitch type was preliminarily conveyed to participants. Participants' hitting movement was recorded by an optical motion capture system and force platform. We calculated timing error based on the difference between the measured impact location (ball position relative to the batter's body at ball-bat impact) and optimal impact location. The timing error between n-th pitch type, (n-1)-th pitch, and the presence or absence of advance information about pitch type (open vs random condition) were analyzed using three-way repeated ANOVA. The results showed that the (n-1)-th pitch type did not affect the timing of impact (p = 0.338). In contrast, the timing errors in open conditions were fewer compared to random conditions (p < 0.001). These results indicate that the pitch type sequence has insignificant effects, and advance information about pitches affects the timing errors. Therefore, having two or more pitch types, reducing the fluctuation of the pitching motion, and the early trajectory of the ball between different pitches potentially lead to increase timing errors.