基于改进的加权累积差分法的人体行为识别*

针对两帧差分法和三帧差分法难以提取到完整的运动剪影,本文提出了一种基于改进的加权累计差分法的人体行为识别方法。通过使用改进的加权累计差分法能通过计算帧的相似度,用于对权值进行自适应变化,从而提取到较为完整的人体运动剪影,然后采用提出的关键帧的模板选取方法和分块特征提取来进行行为的特征提取,最后利用支持向量机构造分类器进行识别。实验结果表明采用改进的加权累积差分法能有效提高人体行为识别率。

Human Behavior Recognition Based on Improved Weighted Accumulative Frame Difference Method

For two frame difference method and three frame difference method is difficult to extract the complete motion silhouettes, this paper presents a method based on improved weighted accumulative frame difference method for human behavior recognition. By using the improved weighted accumulated frame difference method to compute the similarity of frames for adaptive weights change, a relatively complete human movement silhouette can be extracted, and then use proposed key frame template selected method and block features extraction method to extract behavior features. Last, support vector machine is used to construct classifiers for recognition. Experimental results indicate that this improved weighted accumulative frame difference method can be used to improve human behavior recognition rate.