3D Pooling on Local Space-time Features for Human Action Recognition

Successful approaches use local space-timefeatures for human action recognition task including handdesigned features or learned features. However these methodsneed a wise technique to encode local features to make a globalrepresentation for video. For this, some methods use K-meansvector quantization to histogram each video as a bag of word.Pooling is a way used for global representation of an image. Thismethod pools the local image feature over some imageneighborhood. In this paper we extend pooling method called 3Dpooling for global representation of video. 3D pooling representseach video by concatenating pooled feature vectors achieved from8 equal regions of video. We also applied stacked convolutionalISA as local feature extractor. We evaluated our method on KTHdata set and got our best result using max pooling. It improvesthe performance of highly demanded earlier methods.

کلیدواژه ها:

action recognition ، local feature ، independent subspace analysis(ISA) ، pooling