Cost effective trajectory tracking for sports performance enhancement
Sri Lankan athletes and players take part in many sports today, not only locally but also on the international stage. They need to continually improve their performances by having their limb movements monitored, for which the naked eye is just not enough. The high cost of technology on the other hand, is a severe disadvantage for a developing country, which also has to prioritize investments. In Sri Lanka, while products such as Hawk-Eye and Virtual Eye are used in international cricket, technology is seldom used in other sports.
Figure 1 – The initial template (barbell)
Dr Amalka Pinidiyaarachchi from the Department of Statistics and Computer Science, Faculty of Science at the University of Peradeniya launched a project to develop an automated tool for performance evaluation in sports using trajectory analysis, with a focus on weightlifting. Gihan Kuruppu, a weightlifter himself, was her MPhil student at the university’s Postgraduate Institute of Science. The National Research Council of Sri Lanka sponsored the study.
Templates and tests
The complexity of the real-time tracking makes the accuracy of these processes very low. The main problem in real-time trajectory tracking is the dynamic background of the environment. Accuracy can be improved by introducing efficient computer vision techniques. This research project aimed to deliver cost-effectiveness too.
Figure 2- The captured weightlifting trajectory
Although the research considered tracking and trajectory analysis techniques for a range of sports image sequences, it was focused mainly on the weightlifting movements of an athlete. In the snatch lift technique for example, the weightlifting bar needs to be dragged from the floor to its top position in six phases. The “point of interest” that is tracked is the end of the barbell. To obtain the best results, the camera angle is kept perpendicular to the weightlifting bar while capturing the video sequences. Figure 1 shows what is called a template that includes the end of the barbell. It is this template (obtained while originally at rest) that is tracked, by obtaining the best match between sequential template images from one video camera frame to another. This is called the constant template method. Alternatively, the template can be changed from the original one to one that defines the best matched area in the next camera frame and so on – this is called the variable template method. A combination of the constant and variable template methods can also be used.
The comparison of template creation methods was explored using both normally captured data (30 frames per second) and data captured in slow motion (100 frames per second) separately. For the normally captured data, the combined constant and variable template matching method was found to have the highest accuracy. For slow-motion data, all three template creation methods were found to be equally good.
Template matching is done by selecting the area of the current frame that is closest to the template image determined at the previous frame. Such matching is based on statistical measures or tests. Here too there are many measures that can be used. One of the requirements of such measures is that the computational time required to employ it must also be small enough in order to obtain the trajectory as close to “real time” as possible. The “squared difference” criterion was the one with the fastest computational speed.
Once a weightlifter’s trajectory is captured (Figure 2), it can be compared with those of weightlifting champions (which are publicly available). In this project, the trajectories produced by two such experts, publicly available in video form, were combined in order to obtain an optimal trajectory. Weightlifting coaches can then assist athletes to change their actions so that their trajectories get closer to those of the champions.
Cost effective technology
The benefit of this technique is that it only needs a computer with a capture card and a normal video camera. Unlike other commercial systems (which would typically cost over Rs 4 million), this tool needs no attachments fixed to the bar or player and is hence a less invasive method; and can be used in real-time competitions too. The software implemented is also able to obtain other information such as the bar travel distance, total time for a lift, maximum height, maximum velocity, first pull speed and second pull speed. This information too can be used to compare the athlete’s performance with those of champions. The tool can also be used in other sports such as rugby, boxing and athletics.The software works on any Microsoft Windows platform.
Multiple outcomes
The data used in the study were collected from Kurunegala, Kandy and Colombo at weightlifting national level gymnasiums with national level Sri Lankan players. With completion of this research however, a workshop held at the University of Peradeniya generated much interest from the Weightlifting Federation, weightlifting clubs and the Sports Ministry.As a result, this tool is currently used in the Ministry of Sports and Weightlifting Federation of Sri Lanka, to improve techniques for national-level weightlifters. All the national weightlifting competitions use this tool to validate athlete performance. In addition, the tool can also be used in the field of medicine for rehabilitation and orthopaedic re-correcting processes, as it is at the Sports Medicine Unit of the General Hospital in Peradeniya.
The research also won a Nature’s Secrets Research Award in 2013 for conducting commercially viable research. The system is being improved by adding features that are seen in Hawk’s eye, currently used in cricket.
This research was funded by the National Research Council under grant IDG 11-072, and awarded to Dr Amalka Pinidiyaarachchi of the University of Peradeniya. Gihan Kuruppu, the research assistant on this project, is himself a weightlifter. Prof. Saluka Kodituwakku from the Faculty of Science also acted as a co-supervisor for the project. Prof. Vajira Weerasinghe and Dr. Asela Ratnayake from the Faculty of Medicine are acknowledged for their contributions. Asitha Jayawardena was commissioned to write this article as part of the science publicity programme of the National Academy of Sciences of Sri Lanka.