使用Kinect深度图像的静态人体动作识别

Static human motion recognition by using Kinect depth image

为解决传统的使用视频彩色图像序列的智能监控容易受光照、颜色等因素影响的问题,提出结合Kinect深度图像和支持向量机的人体动作识别方法.利用Kinect在监控区域获得实时深度图像,并进行背景擦除,滤波处理和提取Haar特征.通过使用支持向量机的分类算法生成分类器,并对一组特定的静态动作识别结果进行分析.研究结果表明:使用深度图像对于静态动作有较好的识别率,并且与传统的基于彩色图像的智能监控相比,该方法不仅对于光照、颜色等因素不敏感,而且在识别的准确率和效率上均有提升.

In order to solve the problem that the traditional intelligent surveillance with video image sequence are easily influenced by light, color and other factors, this paper proposed a new method for human motion recognition combined Kinect depth image with the support of vector machine. Kinect was used to obtain real-time depth images in the monitored area, and to erase background, filter and extract Haar feature of the images. This study utilized the classification algorithm with support vector machine to generate a classifier, and analyzed the recognition result of a set of specific static human motions. Experimental results show that this method is not only insensitive to light, color and other factors, but also improves the recognition rate and efficiency. The proposed method has high application value.