摘要
针对复杂的交互行为在识别过程中特征维数高、信息冗余大导致的识别率较低的问题,文章提出了基于深度图像和RGB图像特征的低秩稀疏融合算法.采用基于稀疏表示的空间金字塔模型对静态特征和动态特征进行编码融合.最后采用低秩稀疏优化方法进行行为识别,该方法在公共数据集CAD-60和MSR Action Pairs上进行实验验证,取得了理想的效果.
A low-rank sparse fusion algorithm based on depth image and RGB image features was proposed to solve the problem of low recognition rate caused by high feature dimension and information redundancy in the recognition process of complex interactive behaviors.The spatial pyramid model based on sparse representation was used to code and fuse the static and dynamic features.Finally,the low-rank sparse optimization method was adopted for behavior recognition,which was verified by experiments on common data sets CAD-60 and MSR Action Pairs,and the ideal effect was achieved.
作者
马欣
李建军
MA Xin;LI Jianjun(Information Engineering School,Inner Mongolia University of Science and Technology,Baotou 014010,China)
出处
《内蒙古科技大学学报》
CAS
2021年第4期375-381,共7页
Journal of Inner Mongolia University of Science and Technology
基金
国家自然科学基金资助项目(62066036)
内蒙古自治区自然科学基金资助项目(2018MS06018)
内蒙古自治区杰青培育基金资助项目(2018JQ02).
关键词
交互行为识别
特征融合
稀疏
低秩
interactive human action behaviors
feature fusion
sparse
low-rank
