A Generic Virtual Template for Camera Calibration in Soccer Broadcast Footage

Camera calibration is an underlying process for pivotal sports analytics tasks such as augmented reality, player tracking, and scene reconstruction. Camera calibration in soccer aims to estimate the geometrical relationship between the pitch and broadcast footage, however, the international pitch guidelines allow for a variance of up to \(1850\,\text {m}^2 \) in physical soccer pitch size. The objective of this paper is to determine if a generic virtual template can be used as the calibration object for camera calibration in all soccer broadcast footage from internationally approved pitches. An experiment is therefore conducted to determine if the Federation Internationale de Football Association (FIFA) recommended-sized pitch can be morphed into any internationally approved pitch. Four extreme soccer pitches are created in a virtual space to resemble the minimum, maximum, min-max, and max-min pitch dimensions accordingly. The direct linear transform is employed to find a homography between the generic template and extreme image planes. It is concluded that the generic template is well grounded for accuracies up to at least \(96\%\) , calculated with respect to the camera’s perspective. This accuracy method, however, considers the total area of the field and not different segments of the playing field within that area. Therefore, in addition to the standard metric used in the literature, a novel metric is proposed to calculate the combined intersection over union (IoU) as the average IoU per field segment in the visible plane.