出版社：University of Malaya * Faculty of Computer Science and Information Technology
摘要：The human action classification task is a widely researched topic and is still an open problem. Many stateofthe arts approaches involve the usage of bagofvideowords with spatiotemporal local features to construct characterizations for human actions. In order to improve beyond this standard approach, we investigate the usage of cooccurrences between local features. We propose the usage of cooccurrences information to characterize human actions. A tradeoff factor is used to define an optimal tradeoff between vocabulary size and classification rate. Next, a spatiotemporal cooccurrence technique is applied to extract cooccurrence information between labeled local features. Novel characterizations for human actions are then constructed. These include a vector quantized correlogramelements vector, a highly discriminative PCA (Principal Components Analysis) cooccurrence vector and a Haralick texture vector. Multichannel kernel SVM (support vector machine) is utilized for classification. For evaluation, the well known KTH as well as the UCFSports action datasets are used. We obtained stateofthearts classification performance. We also demonstrated that we are able to fully utilize cooccurrence information, and improve the standard bagofvideowords approach.
关键词：local features; human action; classification; spatiotemporal cooccurrence
