Fig 7 - uploaded by Mihnea Jurca
Content may be subject to copyright.
Context in source publication
Context 1
... we are going to cover is when a part of the center circle and a part of the big box can be seen. The four points that are going to be computed are: the right or left extremity of the center circle, the top left or right point of the big box, depending on where the frame is positioned and the two projections of these point onto the top lines (see Fig. 7). The first step is to compute the extremities of the hull contour of the center circle and also to compute the bounding box around it, in order to eliminate the noise lines for the next ...
Similar publications
Citations
... Automatic team affiliation completion is crucial for football game reconstruction [1], which significantly impacts tasks such as player re-identification [7], [20], player tracking [7], [21], and event detection [22]. Traditional research methods [14], [15], [17], [18], commonly employ color histograms or color moments for feature representation and apply k-means clustering for classification purposes. However, relying on a single color feature makes the results highly sensitive to changes in lighting, and the hard clustering approach of kmeans performs poorly in clustering a few targets on the football field, such as goalkeepers and referees. ...
... An unsupervised member classification method [19] was proposed to reduce the dependency on datasets, yet it still necessitates prior labeling of referees. [18] discussed the usability, advantages, and disadvantages of existing supervised and unsupervised algorithms in team affiliation tasks. An unsupervised method combining Convolutional Autoencoders (CAEs) and k-means [17] effectively distinguishes latent representations, achieving good results. ...
... These comparative results are detailed in Table 1, while the qualitative outcomes are illustrated in Fig. 13. Our method generally performs well compared to those cited in [15], [17], [18], particularly on the Sn-gamestate dataset, which presents challenging uneven lighting conditions. Despite these challenges, our method achieves a robust ARI of 0.768, a V-Measure of 0.819, an accuracy of 0.854, and an F1 score of 0.673, which are competitive with the average performance of the referenced methods. ...
In football match videos, team affiliation is typically identified using unsupervised methods, which distinguish individuals based on unique features. These methods reduce the effort needed for dataset labeling compared to supervised approaches. However, uneven lighting in outdoor football scenes often compromises accuracy. This paper introduces a clustering method leveraging color segmentation combined with illumination equalization to address issues such as large shadows and unknown uniform designs. This method distributes personnel information—distinguishing team A, team B, goalkeepers, and referees—relying solely on color features to achieve precise clustering. Compared to established unsupervised methods, our approach demonstrated superior performance on benchmarks including the Sn-gamestate and Soccernet-Tracing datasets, which contain 81, 000 images. Additionally, we developed a shadow correction and color enhancement technique tailored for unevenly lit football scenes. Experimental results show that this method significantly improves clustering accuracy in challenging lighting conditions, boosting the Adjusted Rand Index (ARI) by at least 0.2 and enhancing color restoration markedly.
