基于HHMM的多线索融合和事件推理方法

Cue fusion and event inference based on HHMM

为了解决基于内容检索技术中低层特征与高层语义之间存在语义间隔问题提出了基于多层次线索与事件的分层模型,以及相应的基于分层隐Markov模型(HHMM)的多线索融合和事件推理方法。其中线索是对事件进行推理的要素,它是低层特征与事件之间的中间层次。在将视频流分割为镜头后,从各个镜头中抽取若干与事件密切相关的线索,构造并训练各事件的HMM模型,用于融合线索和进行事件推理。由于输入视频通常包含多个事件,不可避免会遇到时域分割问题,因此构造一个HHMM模型用于同时进行视频流的合理分割和事件的识别。对足球视频的大量实验结果表明,该方法可有效地检测足球视频事件,并在抽取的线索不完全可靠的情况下具有一定的鲁棒性。

A cues fusion and events inference method was developed based on the hierarchical hidden Markov model （HHMM） to bridge the semantic gap between the low-level features and the high-level semantics in content-based retrievals. Cues are introduced into the system as an element for inferring higher-level events. In the system framework, the input video stream is first segmented into shots, then, semantic cues are extracted from the shots based on low-level features, and, HHMM models are built and trained to infer the events from the cues. The input video streams usually contain more than one event, So a temporal segmenting video stream is used to segment events for the HHMM-hased events inference. An HHMM model was developed to group shots and to recognize simultaneously events in a soccer video. Tests on the soccer videos show that the system is effective and robust in inferring events from roughly extracted cues.