Action Recognition Using Low-Rank Sparse Representation

• DOI: 10.1587/transinf.2017edl8176
• Publication Date: 2018-03-01T22:25:40Z
• Authors: Shilei CHENG; Song GU; Maoquan YE; Mei XIE
• Keywords: Representation; Rank (graph theory); Action Recognition; Feature (linguistics); Feature vector

Human action recognition in videos draws huge research interests computer vision. The Bag-of-Word model is quite commonly used to obtain the video level representations, however, BoW roughly assigns each feature vector its nearest visual word and collection of unordered words ignores interest points' spatial information, inevitably causing nontrivial quantization errors impairing improvements on classification rates. To address these drawbacks, we propose an approach for by encoding spatio-temporal log Euclidean covariance matrix (ST-LECM) features within low-rank sparse representation framework. Motivated low rank recovery, local descriptors a temporal neighborhood have similar should be approximately rank. learned coefficients can not only capture global data structures, but also preserve consistent. Experimental results showed that proposed yields excellent performance synthetic datasets are robust variability, view variations partial occlusion.