基于密度轨迹与句法规则的复杂行为识别

Recognition of Complex Behavior Using Density Track and Syntax Rules

人体行为识别是计算机视觉研究中重要的领域.为了识别视频流中复杂的人体行为,提出一种新的框架.特征提取上,提取了轨迹形状描述符、SURF(结构描述符)和HOF(运动描述符)作为特征描述符.识别上采用了先简单后复杂的行为识别过程,首先利用HMM对简单的行为建模,然后根据句法规则描述复杂行为并进行识别.为了验证文中所提方法的有效性,在公共数据库上对该法进行了验证,并与其他方法进行了对比实验,结果表明所提的方法在识别复杂行为上很有效.

Human behavior recognition is an important areas of computer vision research. A newframework is proposed to recognize complex human behavior in the video streams. As for feature extraction,use track shape descriptors,SURF（ structure descriptor） and HOF（ motion descriptor） as feature descriptors. And for recognition use a method first for simple behavior recognition then for complex behavior recognition,first use HMMfor simple behavior modeling,then describe the complex behavior based on syntactic rules and recognition. To verify the validity of the proposed method,experiment on public database compared with other methods,the results showthat the method is effective in complex behavior recognition.