AAAI Publications, The Twenty-Seventh International Flairs Conference

Font Size: 
Action Classification Using Sequence Alignment and Shape Context
Sultan Almotairi, Eraldo Ribeiro

Last modified: 2014-05-03


In this paper, we describe a method for classifying human actions from videos. The method uses the Longest Common Sub-Sequence (LCSS) algorithm to match  actions represented by sequences of pose silhouettes. The silhouettes are extracted from each video frame using foreground segmentation. The main novelty of our method is the use of the Shape Context (SC) and Inner-Distance Shape Context (IDSC) as a pairwise shape-similarity measurement for constructing the sequence-alignment cost matrix. Experiments performed on two action datasets compare our approach favorably with previous related methods.


Shape Context; Human Motion Recognition; Longest Common Subsequence; Dynamic Time Warping; Inner-Distance Shape Context.

Full Text: PDF