Validation of a video-based system for automatic tracking of tennis players

The purpose of this study was to validate an automatic method for tracking tennis players. The method is based on algorithms for camera calibration, image pre-processing, segmentation, filtering, tracking and computer vision tools. Two digital video cameras (30 Hz) were used to register player movement on a tennis court. The validation of the 2D-reconstructed points was based on the following analysis: intra-operator repeatability (.009 m), inter-operator repeatability (.007 m), relative error (.24% for length and .52% width coordinates) and ICC (.88 consistency and .89 absolute agreement, p < .05). The mean bias and precision were .33 and .15 m, respectively, resulting in an accuracy of .36 m. There were significant linear regressions (p < .05) between positions obtained by the automatic and manual tracking procedures (R2 = .9903 for x-coordinate, R2 = .9952 y-coordinate). The RMS error comparing both methods was .51 m. The percentage of automatic tracking (when the system identified player’s position correctly, without operator intervention) during a simulated match reached 99.98% of the 22,000 frames processed. The distances covered by the two players during a set were 1229.6 and 1083.1 m. In conclusion, the method proposed revealed to be simple, valid and consistent to track tennis players on the court.