Comparison of Different Template Matching Algorithms in High Speed Sports Motion Tracking Gihan Kuruppu, Chandrabose Manoj, S. R. Kodituwakku and U. A. J. Pinidiyaarachchi Abstract-Tracking of sports motion is a challenging task. This paper presents a comparison of different template matching methods that can be used in such motion tracking applications. Six methods were tested with dynamic and still background conditions. Their performance is analyzed using different video sequences obtained by considering sports such as table tennis, weightlifting, 100m athletics and high jump. Comparison is carried out by means of statistical methods to compare the six different template matching algorithms with the ground truth in terms of accuracy and processing time. The results indicate that, even though there are no statistically significant differences among different template matching algorithms, four out of the six methods have perfect matches with ground truth with respect to the accuracy. Among the six methods considered, the squared difference method is found to be the best in terms of processing time. Index Terms—Sports Motion Tracking, Object Tracking, Fast Motion Tracking, Template Matching Algorithms, ANOVA, Kruskal-Wallis . I. INTRODUCTION The invention of computers has greatly modernized and improved the world of sports. Computer technology allows for increased accuracy in different sporting events and can be used to correct an athlete’s bio-mechanic techniques which are relatively difficult to evaluate using the naked eye. A sport is a competitive physical activity which contributes to the improvement of physical, social and psychological aspects of life in addition to providing an entertainment to the participants and spectators alike. The competitiveness of sports is ever increasing and every professional athlete’s ultimate aim is to win an international event like an Olympic gold medal. Therefore, athletes try their very best to improve their techniques in order to achieve this goal. The honing of an athlete’s technique can be achieved with computer technology by tracking, in real time, an athlete’s movement using image processing algorithms in order to correct or improve their bio-mechanical techniques thereby improving their performance [1]. In selecting an algorithm for real time movement tracking, two factors are taken into consideration: 1. Accuracy of the algorithm 2. Processing time of the algorithms Image processing tracking algorithms use complex searching techniques and in most of the cases tracking algorithms are used in real time. Real time processing requires powerful processing and optimized algorithms. The difficult task here is to accurately track high-speed motion in real time instances like cricket ball tracking, weightlifting bar motion tracking as opposed to certain other instances of high- speed motion tracking where the accuracy of the trajectory is not important (vehicle tracking). When the field of sports is considered, the calculated motion trajectory must be accurate enough to correct athletes’ techniques. Sports are mainly divided into two categories as indoor and outdoor sports. The major difference with regard to selecting an image tracking algorithm is the fact that most indoor games take place in a still background as opposed to outdoor games that take place in a constantly changing environment (Fig 1). Thus selecting an algorithm for sports’ motion tracking is not a trivial task. In this work, six template matching methods, namely, squared difference, Normalized squared difference, Cross correlation, Normalized cross correlation, Correlation coefficient and Normalized correlation coefficient that can be used in high speed motion tracking are tested and evaluated. The rest of this paper is organized as follows. Section II presents the related work. Section III discusses sports motion tracking and the methodology applied in the study. Section IV describes the results obtained. Finally Section V concludes the paper with a discussion. Fig. 1. a) Still Background TTB moving b) 100m Athlete body movement Gihan Kurppu and Chandrabose Manoj are with the Postgraduate Institute of Science, University of Peradeniya, Sri Lanka. U. A. J. Pinidiyaarachchi and S. R. Kodituwakku are with the Department of Statistics and Computer Science, Faculty of Science, University of Peradeniya, Sri Lanka.