摘要
由于传统图像序列识别方法受噪声因素影响,导致序列识别精度较低,提出一种基于低秩分解的异常步态活动图像序列识别方法。设定步态历史图像序列作为标准图像序列,根据矩函数的特征向量,列出Zermike矩提取图像序列特征向量数据。对步态图像像素点矩函数特征进行识别并转化为向量格式,利用低秩分解方法构建结构化矩阵低秩表示模型,去除序列特征向量数据噪声。对分解去噪后的数据进行Curvelet特征转化,得到形变约束完成图像序列识别。仿真结果表明,所提方法的图像序列识别率达到了90%,充分说明所提方法的识别精度较高,且去噪效果十分理想,优于现有方法。
Because the traditional method is affected by noise, the accuracy of sequence recognition is low. Therefore, a method to recognize abnormal gait active image sequence based on low rank decomposition was put forward. Firstly, the historical image sequence of gait was set as the standard image sequence. According to the feature vector of moment function, the feature vector data of image sequence extracted by Zermike moment were listed. The moment function features of pixels in gait image were recognized and transformed into the vector format. The low-rank representation model of structured matrix was built based on low-rank decomposition method, and the noise of sequence feature vector data was removed. In addition, Curvelet feature transformation was conducted on the data after decomposition and denoising, and then the deformation constraints were obtained, and thus to complete the image sequence recognition. Following conclusions can be drawn from simulation results: the image sequence recognition rate of the proposed method is more than 90%, so the proposed method has higher recognition accuracy and more ideal denoising effect.
作者
邱妍妍
高增
QIU Yan-yan;GAO Zeng(Longqiao College of Lanzhou University of Finance and Economics,Lanzhou,Gansu,730000,China;Faculty of Sport Science and Coaching,Universiti Pendidikan Sultan Idris,Tanjong Malim,Malaysia 35900)
出处
《计算机仿真》
北大核心
2021年第6期415-418,共4页
Computer Simulation
关键词
低秩分解
异常步态
图像序列
识别
多特征
Low-rank decomposition
Abnormal gait
Image sequence
Identification
Multi-feature
