gms | German Medical Science

Introduction: We regard sport in general as one aspect of an all-embracing healthy lifestyle and we aim at using computer science techniques to support training and competition. As the requirements vary according to the type of sport [1], [2], [3], [4], a structured training management system specifically for each type of sport is desired. In this work, we focus on the sport of squash due to its favorable constraints such as the compact playing field of about 62m² and the fact that there are usually only two athletes on the court at the same time. We measure and analyze the players’ stress and strain parameters in order to support the training process in a controlled and targeted manner.

Method: To measure physiological strain, we have developed a camera-based tracking system using computer vision techniques that allows us to calculate distances travelled on court by each player separately [5]. Contrary to other approaches which use a ceiling mounted camera [6], we use a static camera located behind the court which also serves as a calibration object. Our algorithm uses mixture-based Gaussian background subtraction with clustering for player detection and dominant color matching for identification.

The stress parameter is measured by using a chest strap (Zephyr Bioharness), worn continuously during the entire match. It writes reliable [7] heart rate (HR) measurements on an internal storage. The belt also contains an integrated acceleration sensor which is used to synchronize the HR with the video and thus with our tracking data.

After synchronization, the resulting data is then classified into rallies and breaks in-between rallies by manual scoring annotation. Afterwards the classified data parts are integrated into a relational database schema. During the integration, our data is further enriched with meta-information such as linear regression parameters for HR during individual phases. This allows us to query, filter and evaluate the recorded data up to the level of individual rallies.

Results: Currently, our dataset consists of two matches lasting 1593 and 1602 seconds (s). In total they consist of eight games [347s±186s] (mean±SD) with a total amount of 149 rallies [6.9s±4.4s]. Since we have two players for each rally, our rally table contains a total of 298 entries. The players’ covered distance during rallies is [348m±91.96m] per game. The pearson correlation (PCC) score for rally duration versus covered distance is 0.72. If we consider winning rallies only, a PCC for the rally's preceding HR regression gradient and covered distance shows a negative relation of -0.28, whereas for losing rallies the PCC is 0.03.

Discussion: Our current dataset with n=149 rallies is small and therefore does not allow any significant conclusions to be drawn about the interrelationship regarding the stress-strain model. However, initial analysis showed indications for further investigations. In addition to the necessary database expansion, an inclusion of further parameters such as respiration rate, is planned. This hopefully will lead us to a squash sport specific model for optimizing training on the one hand and provide assistance for a recreational healthy gameplay on the other.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.