融合归一化语义加权和规则的足球视频进球检测

Soccer Goal Detection with Fusion Normalized Semantic Weighted Sum Rules

针对足球视频精彩进球事件检测,提出一种归一化的语义加权和规则足球进球检测融合方案.首先构建了进球事件的隐马尔科夫模型(HMM);然后提出一种语义观测权重的镜头新特征,以此建立归一化的语义加权和规则,分别实现了基于HMM方法和语义加权和规则方法的进球事件检测;最后提出一种基于逻辑距离的融合方案,将2种方法的检测结果通过最优权重进行决策级融合,显著地提高了进球事件的检测性能.采用文中方案建立的语义加权和规则基于客观的视频统计信息、不过多依赖于人的主观观察,克服了同类方法中的人力耗费问题,不需要复杂训练,计算量较小;并通过实验证明了该方案的有效性.

A normalized semantic weighted sum soccer goal detection fusion scheme is proposed in this paper. First, the hidden Markov model （HMM） for the goal event is constructed. Second, a new feature of semantic observation weights for the shots is proposed. With this new feature, a new manual rule, named normalized semantic weighted sum rule, is established. The goal event detection in soccer videos is implemented based on the HMM method and the normalized semantic weighted sum rule method, respectively. Finally, a fusion scheme based on the logic distance is proposed, which carries out the decision-level fusion with the optimal weights to fuse the HMM and the rule detection results, and significantly improves the performance of the goal event detection. The normalized semantic weighted sum rule proposed in this scheme uses of the statistical video information objectively, relies less on manual subjective observations, and overcomes the huge manual efforts in existing methods. It does not require complex training and the computational cost are relatively small. The experiments validate of effectiveness of this scheme.