融合HCRF和AAM的足球视频精彩事件检测

Fusion of HCRF and AAM Highlight Events Detection in Soccer Videos

精彩事件检测在体育视频语义分析领域具有很高的学术研究价值和广泛的市场应用前景.利用隐条件随机场(hidden conditional random field,HCRF)模型在表达和识别语义事件方面的强大功能,创新性地提出了一种融合了HCRF和情感激励模型(affective arousal model,AAM)的精彩事件检测方法.首先,通过精彩事件视频结构语义分析,定义了13种多模态语义线索,以准确描述精彩事件富含的语义信息;其次,在基于概念格的多模态语义线索聚类基础上,添加时域特征信息,以构建特征值加权的情感激励模型,得到了各类精彩事件的情感激励值;最后,在小规模训练样本情况下,有效建立了各类精彩事件检测的HCRF模型,基于视频语义镜头序列、情感激励值序列和精彩事件之间的映射关系,从多模态语义线索、视频结构语义、情感语义等多个维度挖掘了精彩事件的潜在规律,实现了同一HCRF模型下各类精彩事件的同时检测.实验证明了该方法的有效性.

Highlight event detection in soccer videos has high academic research value and wide market application prospect in the field of sport video semantic analysis. Based on the powerful expression o{ hidden conditional random field （HCRF） model in the expression and identification of semantic event, a fusion HCRF and affective arousal model （AAM） framework for highlight event detection is put forward. Firstly, through the analysis of the structural semantics of the wonderful event video, thirteen kinds of multi-modal semantic clues are defined to accurately describe the included semantic information of the wonderful events. Secondly, on the clustering foundation of the multi-modal semantic clues by concept lattice, time-domain features are added to establish an affective arousal model based on feature weight coefficient, and then the affective arousal value of the different kinds of highlights events is calculated. Finally, the above observed sequence is used as HCRF model input in the case of small-scale training samples, and a wonderful event detection HCRF model is effectively established based on the mapping relationship between the sequences of video semantic shots, affective arousal values and the highlight events. The inherent laws of the wonderful events are excavated from multiple dimensions like multi-modal semantic clues, video structure semantics, and affective semantics. The detection of wonderful events is simultaneously achieved by using the same HCRF model. Experimental results show the effectiveness of this paper.