Extraction of Positional Player Data from Broadcast Soccer Videos

Keypoint-less Camera Calibration for Sports Field Registration in Soccer

“…If the same ground-truth segmentation is used as input for TVCalib and [6] (and its variants [34]), TVCalib achieves superior results when evaluating both, the calibration task and the homography estimation.…”

“…Comparison to State of the Art: As the majority of approaches estimate homography matrices, it is reasonable to apply decomposition on both (1) predicted matrices ( Ĥ) or (2) already manually annotated or ground-truth matrices (H). More concrete, we have reimplemented the approach from Chen and Little [6] as they rely on synthetic [34]. As the second approach, we apply the official implementation from Jiang et al [23] for homography estimation denoted as Ĥ( [23]).…”

“…SN-Calib Dataset: SoccerNetV3-Calibration [10] (SN-Calib) consists of 20 028 images taken from the Soccer-Net [14] videos (500 matches). In contrast to other (private) datasets [27,30,34] or WC14 [20], which cover the main broadcast camera (here denoted as center), the SN-Calib dataset covers more camera locations placed in the stadium. An example setting may consist of two cameras that are placed also on the same tribune as the central broadcast camera, but are closer located to the side lines (main left and right).…”

Evaluating deep tracking models for player tracking in broadcast ice hockey video

“…There are a number of recent studies dealing with player tracking in basketball [19,13,27] and soccer [20,9,21,7]. For basketball player tracking, Sangüesa et al [19] demonstrated that deep features perform better than classical handcrafted features for basketball player tracking.…”

“…Hurault et al [9] introduce a self-supervised detection algorithm to detect small soccer players and track players in non-broadcast settings using a triplet loss trained re-identification mechanism, with embeddings obtained from the detector itself. Theiner et al [21] present a pipeline to extract player position data on the soccer field from video. The player tracking was performed with the help of CenterTrack [29].…”