Oscar Broddegård and Caesar Ringström will present their MSc thesis “Calibration of extrinsic camera parameters by identifying lines in a monocular view of a soccer field” on Wednesday 5th of June, 11:15 in MH:309A.

Soccer is witnessing technological advancements in a variety of areas, of which many are dependent on camera calibration. This thesis addresses the challenge of calibrating the extrinsic parameters of cameras used in soccer broadcasts using a monocular view of the field. Accurate camera calibration is essential for various applications such as player positioning, field modelling, and improving the overall broadcast quality. This research focuses on evaluating techniques for detecting lines on the soccer field and using these lines to estimate the camera’s position and orientation, and the field’s aspect ratio. The study explores multiple line detection methods, including traditional line detection techniques like Hough Transformation (HT) and Line Segment Detector (LSD), as well as advanced approaches like Holistically Attracted Wireframe Parsing (HAWP) and Semantic Line Detection (SLNet). Each method’s performance is evaluated based on accuracy, robustness, and computational efficiency. The proposed pipeline leverages detected field lines to calibrate the camera by estimating a plane-induced homography, mapping image points to world coordinates. The calibration process is validated through reprojection error analysis, ensuring the detected lines and their intersections provide reliable input for camera pose estimation, and that the assumptions of a planar rectangular field are valid. Results indicate that HAWP outperforms the other tested methods in detecting soccer field lines, but since it leverages detected junctions to find lines it becomes very sensitive to occlusions. While other tested methods show some potential, they require further refinement for consistent performance. Furthermore, the camera calibration algorithm based on the homography shows promising results, suggesting that the field geometry assumptions are valid. Finally, a proposition of further development is presented, providing a foundation for future research within the field.