Semantic analysis based on fusion of audio/visual features for soccer video

In order to analysis the semantic content of soccer video, the audio/visual features are effectively extracted. Based on the theory that the variation of video content would cause the fluctuation of viewers' affection, the highlight time curve (HTC) is generated by fusion of affection arousal factors to reveal the excitement of the game. The semantic boundaries of highlights are determined by HTC combined with the domain knowledge of soccer video. With the help of distinguishable highlight feature vectors (HFVs), highlights are classified into goal, shoot, and foul. Compared with the existing works, the main contributions of this paper are as follows. We proposed a novel Hough transform based whistle detection algorithm and achieves more effective performance. A robust goalmouth detection algorithm is presented and contributed to the highlight classification phase. The highlights with semantic boundaries are accurately extracted and classified. Experiments conducted on real world soccer videos demonstrated the good performance of the proposed framework. (C) 2021 The Authors. Published by Elsevier B.V.