Using synthetic camera poses for camera calibration in soccer videos.
Saved in:
| Title: | Using synthetic camera poses for camera calibration in soccer videos. |
|---|---|
| Authors: | Mavrogiannis, Panagiotis1 (AUTHOR) pmavrogiannis@unipi.gr, Maglogiannis, Ilias1 (AUTHOR) imaglo@unipi.gr |
| Source: | Multimedia Tools & Applications. May2025, Vol. 84 Issue 18, p18967-18991. 25p. |
| Subjects: | Soccer tournaments, Soccer fields, Camera calibration, Image processing, Artificial intelligence |
| Abstract: | Camera calibration is the process of estimating the parameters that describe the viewpoint of a camera. These parameters allow the mapping of points between two 2D planes, with the first plane being an image produced by a camera. In the context of soccer camera calibration, the second plane is the top view of the soccer field, while the image from the camera depicts a screenshot of a game. The points of interest, such as players, referees, and the ball, are mapped from the image as coordinates on the top-view. In this paper, a novel two-stage approach for camera calibration in soccer videos is presented. In the first stage, conducted prior to a soccer game, the physical location of a static camera is identified and a relevant deep CNN model is trained for this location. In the second stage, executed during the actual game, the model estimates the remaining camera parameters per frame, including rotation and lens focus. The second stage concludes with a refining step for the initial estimation. Both stages utilize EfficientNet models that accept images of visual landmarks from the field as input and are trained using fully synthetic data generated by rule-based algorithms. The proposed method is applicable to every camera location within a soccer stadium, with a relatively panoramic viewpoint. A comprehensive review of existing approaches is provided, and the details of the method and the synthetic datasets are discussed. The method is evaluated using known metrics, and the experimental results are compared against other approaches from the literature. Evaluation is conducted using datasets containing images from various camera locations, including the World Cup 2014 dataset, SoccerNet-V2, and the new DFVA dataset. The paper concludes with a discussion of the results, the limitations and applicability of the proposed approach. [ABSTRACT FROM AUTHOR] |
| Copyright of Multimedia Tools & Applications is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
|
Full text is not displayed to guests.
Login for full access.
|
|
| Abstract: | Camera calibration is the process of estimating the parameters that describe the viewpoint of a camera. These parameters allow the mapping of points between two 2D planes, with the first plane being an image produced by a camera. In the context of soccer camera calibration, the second plane is the top view of the soccer field, while the image from the camera depicts a screenshot of a game. The points of interest, such as players, referees, and the ball, are mapped from the image as coordinates on the top-view. In this paper, a novel two-stage approach for camera calibration in soccer videos is presented. In the first stage, conducted prior to a soccer game, the physical location of a static camera is identified and a relevant deep CNN model is trained for this location. In the second stage, executed during the actual game, the model estimates the remaining camera parameters per frame, including rotation and lens focus. The second stage concludes with a refining step for the initial estimation. Both stages utilize EfficientNet models that accept images of visual landmarks from the field as input and are trained using fully synthetic data generated by rule-based algorithms. The proposed method is applicable to every camera location within a soccer stadium, with a relatively panoramic viewpoint. A comprehensive review of existing approaches is provided, and the details of the method and the synthetic datasets are discussed. The method is evaluated using known metrics, and the experimental results are compared against other approaches from the literature. Evaluation is conducted using datasets containing images from various camera locations, including the World Cup 2014 dataset, SoccerNet-V2, and the new DFVA dataset. The paper concludes with a discussion of the results, the limitations and applicability of the proposed approach. [ABSTRACT FROM AUTHOR] |
|---|---|
| ISSN: | 13807501 |
| DOI: | 10.1007/s11042-024-19783-8 |
