This is an in-depth study on the expansion of a tennis ball-tracking system, Trackscore, to function under changing lighting conditions during a tennis match. The research aims to address the challenge of adjusting the camera settings and processing the images captured to maintain a high precision in identifying the tennis ball. Results from this project are promising, demonstrating the potential for a robust and complete system to tackling the current problem of fluctuating lighting conditions.
C++, OpenCV, CUDA
The goal of the project was to extend the utility of the tennis detection system, Trackscore, to outdoor and varied indoor settings, where lighting conditions fluctuate. By first calibrating the hardware (initial calibration) to the scene and having it recalibrated (continuous calibration) when lighting conditions change during the match, we aimed to ensure the system’s adaptability and precision in real-world scenarios, making it accessible and reliable for diverse applications. This has to be done with minimal performance impact.
Trackscore is an automatic linecalling system for tennis, that relies on computer vision to track the ball. This system was originally optimized to work indoors where lighting conditions are consistent. The systems capabilities are challenged in suboptimal lighting condition and when these conditions change over the course of the game.
The results indicate that the system maintains high precision in detecting tennis balls, even when subjected to different environmental settings and varying times of day. Furthermore, our performance analysis shows that the system does not experience any significant loss in functionality and continues to operate as intended in real-time. This demonstrates the robustness of the algorithms we have developed and their ability to adapt to dynamic lighting conditions. However, despite these promising results, there is considerable room for improvement, especially in refining the system's accuracy and expanding its adaptability through more extensive field-testing in diverse and unpredictable scenarios.
Project Duration: 19.02.2024 - 16.08.2024
Effort: 360 hours per person
Team Size: 2
Janic Berger, Christopher Trachsel
Prof. Dr. Christoph Stamm chrisoph.stamm@fhnw.ch
