Classification of Dynamic Sequences of 3D Point Clouds

The subject of this article is 3D action recognition in point cloud sequences. A popular approach to classification of point clouds is the Bag-of-Words method, which classifies histograms of spatial features (as described e. g. by Toldo et al. in "The bag of words approach for retrieval and categorization of 3D objects", 2010). This approach is, however, less effective when applied to action recognition of similar agents (e. g. humans). We will compare a simple HMM-based classifier with the well known Bag-of-Words scheme method, within sensible parameters for 3D point clouds close range acquisition methods. We then show that the dynamic classifier performs better when applied to action recognition of objects of the same type.