Christopher Brumann

Nowadays, video recordings of sport events is standard practice for a variety of applications, ranging from entertainment to competition analysis. Beside that, analysis of athletes while exercising is of particular interest for their coaches in order to gain insight into training quality and allow for training control. To bring together video recordings and the desire for analysis we present the implementation of a genetic algorithm (GA) for the important step of camera calibration. Our implementation can be used not only in a prospective but also in a retrospective manner for the squash sport. We do not rely on directly or manually provided image-world coordinates, but rather only on the playing field as known geometric object, present in the physical camera’s captured image. To find the best GA configuration, we evaluate all combinations of 2 initialization-, 2 fitness-, 3 selection-, 4 crossover-, and 2 mutation strategies. We apply and evaluate the GA’s accuracy on synthetic, artificial renderings, and real world data as well as comparing it to other standard optimization algorithms. Our results reveal the importance of correct camera placement and show sufficient accuracy for our goal of athlete movement analysis. The results will serve for a automatic athlete movement analysis tool to support squash specific training procedures.

In sport science, athlete tracking and motion analysis are essential for monitoring and optimizing training programs, with the goal of increasing success in competition and preventing injury. At present, contact-free, camera-based, multi-athlete detection and tracking have become a reality, mainly due to the advances in machine learning regarding computer vision and, specifically, advances in artificial convolutional neural networks (CNN), used for human pose estimation (HPE-CNN) in image sequences. Sport science in general, as well as coaches and athletes in particular, would greatly benefit from HPE-CNN-based tracking, but the sheer amount of HPE-CNNs available, as well as their complexity, pose a hurdle to the adoption of this new technology. It is unclear how many HPE-CNNs which are available at present are ready to use in out-of-the-box inference to squash, to what extent they allow motion analysis and if detections can easily be used to provide insight to coaches and athletes. Therefore, we conducted a systematic investigation of more than 250 HPE-CNNs. After applying our selection criteria of open-source, pre-trained, state-of-the-art and ready-to-use, five variants of three HPE-CNNs remained, and were evaluated in the context of motion analysis for the racket sport of squash. Specifically, we are interested in detecting player’s feet in videos from a single camera and investigated the detection accuracy of all HPE-CNNs. To that end, we created a ground-truth dataset from publicly available squash videos by developing our own annotation tool and manually labeling frames and events. We present heatmaps, which depict the court floor using a color scale and highlight areas according to the relative time for which a player occupied that location during matchplay. These are used to provide insight into detections. Finally, we created a decision flow chart to help sport scientists, coaches and athletes to decide which HPE-CNN is best for player detection and tracking in a given application scenario.

In this paper, we present a method for locating and tracking players in the game of squash using Gaussian mixture model background subtraction and agglomerative contour clustering from a calibrated single camera view. Furthermore, we describe a method for player re-identification after near total occlusion, based on stored color- and region-descriptors. For camera calibration, no additional pattern is needed, as the squash court itself can serve as a 3D calibration object. In order to exclude non-rally situations from motion analysis, we further classify each video frame into game phases using a multilayer perceptron. By considering a player’s position as well as the current game phase we are able to visualize player-individual motion patterns expressed as court coverage using pseudo colored heat-maps. In total, we analyzed two matches (six games, 1:28h) of high quality commercial videos used in sports broadcasting and compute high resolution (1cm per pixel) heat-maps. 130184 manually labeled frames (game phases and player identification) show an identification correctness of 79.28±8.99% (mean±std). Game phase classification is correct in 60.87±7.62% and the heat-map visualization correctness is 72.47±7.27%.