基于Petri网推理的监控视频事件识别

Petri网模型对复合时序事件具有较好的描述和检测能力,提出基于Petri网描述和推理的监控视频事件信息的自动识别方法。将运动目标及其特征间的空间关系用Petri网的库表示,时间关系及其他推理规则用变迁表示,结合低级计算机视觉算法获得的运动目标特征以及基本事件信息,通过对Petri网推理执行,实现监控视频复杂语义事件的交互式查询,通过实验验证了该方法的有效性。

Recognition of surverillance video events based on Petri net referring approach

In view of the fairish capability of Petri net for describing and detecting compound temporal events, the paper pro-posed a novel approach for recognizing events information from surveillance video by inferred with Petri net. Space relations between moving objects and their features are represented with places. And temporal relations and other rules are represented with transi-tions. Semantic queries about compound video events are mapped into a Petri net, which is inferred with interpretations of moving objects detected by some low-level computer vision algorithms. Example is given for illustrating the principle and shows that user queries about complex semantic events are effectively recognized.