During the last decade, performance improvements in top sports have been increasingly driven by technological innovations. This paper discusses the application of video analysis for training registration in swimming. In current practice, coaches have limited means to evaluate objectively and quantitatively how a training session was carried out. We propose the use of a video-based registration system in order to help the coach in acquiring such information. The system uses multiple cameras to cover the swimming pool. By using a simple background modeling and blob tracking method swimmers are tracked and their lap times are estimated. The main limitation of the system is the failure to detect swimmers at the pool ends while they are resting or underwater. This can lead to the necessity to perform manual interactions to associate laps to swimmers and a systematic underestimation of the lap times of typically 1.5 seconds. Our results can be used to formulate training guidelines that can help overcome some limitations of the system so that, with little or no manual effort, the system could be used in practice to do quantitative measurements. The information on the actual training performance of the swimmers could then be compared with the training schedule made beforehand and used to further optimize the training program.
© Copyright 2007 International Journal of Computer Science in Sport. Sciendo. All rights reserved.