基于累积运动能量图像的人体动作识别

Recognizing Human Actions Using Accumulative Motion Energy Image

指出了动作识别中的最大困难是难以提取有效的特征来准确描述人体的动作,动作模板是众多方法中的一种简单有效的方法,用来描述动作特征的经典动作模板是运动历史图像.由于受噪声的干扰,用运动历史图像描述复杂环境下的人体动作并不十分理想.为了得到比运动历史图像更加有效的动作模板,提出了将视频序列中的运动能量信息用一张图描述出来,称之为累积运动能量图像,提取其直方图特征来表征人体动作.经You Tube数据集上的实验表明:该累积运动能量图像的识别率比同类方法高.

This paper points that very difficult to extract effective feature for describing the action of the human body in action recognition. Action template is a simple and effective method to describe the motion characteristics, in which motion history image is classic method. It isn′t satisfactory using motion history image to describe the human action in complex environment due to noise. In order to obtain more effective action template, this paper presents to use an image to describe motion energy in the video sequence, which is called accumulative motion energy image. The histograms extracted from the image are used to represent human action. Experiments on YouTube datasets show that accumulative motion energy image exceed to similar methods in recognition rate.