A Universal Protocol to Benchmark Camera Calibration for Sports

A Universal Protocol to Benchmark Camera Calibration for Sports

15 Apr 2024 | Floriane Magera, Thomas Hoyoux, Olivier Barnich, Marc Van Droogenbroeck
The paper introduces a new benchmarking protocol, named ProCC, for camera calibration in sports analytics. The current benchmarks, primarily based on homography annotations, are limited to sports field registration and do not fully capture the broader capabilities of camera calibration. ProCC aims to address these limitations by being agnostic to the camera model and evaluating camera calibration methods using the reprojection of arbitrary 3D objects. The protocol uses semantic point annotations of sports field markings as ground truth, which are valid for any camera type. The evaluation metric, JaC, is defined to measure the proportion of correctly imaged field elements, providing a more intuitive and comprehensive assessment of camera calibration quality. Experiments on datasets such as the World Cup 2014, CARWC, and SoccerNet demonstrate the effectiveness of ProCC in evaluating camera calibration methods, showing that richer camera models, including radial distortion, yield better results. The paper also highlights the importance of selecting appropriate camera models for accurate player tracking and discusses the limitations and future directions of the proposed protocol.The paper introduces a new benchmarking protocol, named ProCC, for camera calibration in sports analytics. The current benchmarks, primarily based on homography annotations, are limited to sports field registration and do not fully capture the broader capabilities of camera calibration. ProCC aims to address these limitations by being agnostic to the camera model and evaluating camera calibration methods using the reprojection of arbitrary 3D objects. The protocol uses semantic point annotations of sports field markings as ground truth, which are valid for any camera type. The evaluation metric, JaC, is defined to measure the proportion of correctly imaged field elements, providing a more intuitive and comprehensive assessment of camera calibration quality. Experiments on datasets such as the World Cup 2014, CARWC, and SoccerNet demonstrate the effectiveness of ProCC in evaluating camera calibration methods, showing that richer camera models, including radial distortion, yield better results. The paper also highlights the importance of selecting appropriate camera models for accurate player tracking and discusses the limitations and future directions of the proposed protocol.
Reach us at info@study.space
[slides and audio] A Universal Protocol to Benchmark Camera Calibration for Sports