Subspace Learning for Action Recognition

Recently human action recognition [4, 5, 8–10] has aroused widely attention for public surveillance system, elder service system, etc. However, the data captured by webcams are often high dimensional and usually contain noise and redundancy. So it is crucial to extract the meaning information by mitigating uncertainties for higher accuracy of recognition task. From this motivation, there are two topics we propose in this chapter: (1) select key frames from a video to remove noise and redundancy and (2) learn a subspace for dimensional reduction to reduce time complexity.