Recently, the group of mode recognition and intelligent information processing from College of Information and Control Engineering has new finding on the research of p-Laplacian Regularized sparse coding. The paper p-Laplacian Regularized Sparse Coding for Human Activity Recognition has been published on IEEE Transactions on Industrial Electronics. The research has been a co-work by the researchers from China University of Petroleum and University of Technology Sydney.

With the development of portable intelligent equippments and network technology, tera-scale pictures and Videoes have been put online everyday. The artificial classification of those data is  time-consuming and impossible. Under such context, the automatic analyzing, processing and classifying technology is  eagerly needed. The group proposed a generalized version of Laplacian regularized sparse coding for human activity recognition called p-Laplacian regularized sparse coding (pLSC). The p-Laplacian is a nonlinear generalization of standard graph Laplacian and has tighter isoperimetric inequality. As a result, pLSC provided superior theoretical evidence than standard Laplacian regularized sparse coding with a proper p. It also provided a fast iterative shrinkage-thresholding algorithm for the optimization of pLSC. thesparse codes learned by the pLSC algorithm was input into support vector machines and conduct extensive experiments on the unstructured social activity attribute dataset and human motion database (HMDB51) for human activity recognition. the results showed  that the proposed pLSC algorithm outperforms the manifold regularized sparse coding algorithms including the standard Laplacian regularized sparse coding algorithm with a proper p.

Framework of human activity recognition. First, extracting the representative features of human activity including SIFT, STIP, and MFCC. Then, concatenating the histograms formed by bags of each feature. Third, learning the sparse codes of each sample and the corresponding dictionary simultaneously by the pLSC algorithm. Finally, inputing the learned sparse codes into classifiers, i.e., SVMs to conduct human activity recognition.

In recent years, the group of mode recognition and intelligent information processing has made big progresses on the research of   multimedia computing,  mode recognition and intelligent information processing. The findings have been published on IEEE Trans. on Image Processing, IEEE Trans. on Industrial Electronics, IEEE Trans. on Multimedia, Pattern Recognition, Computer Vision and Image Understanding. Learn more about the paper: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7448894

Editor: Bu Lingduo

Source: UPC News Center