A Novel Compositional Approach for 3D Arm-Hand Action Recognition

TitleA Novel Compositional Approach for 3D Arm-Hand Action Recognition
Publication TypeConference Proceedings
Year of Conference2012
AuthorsGori, I, Fanello SR, Metta G, Odone F
Conference NameIEEE - Conference on Computer Vision and Pattern Recognition - CVPR 2012
VolumeSubmitted
Abstract
In this paper we propose a fast and reliable visionbased framework for 3D arm-hand action modelling, learning and recognition in human-robot interaction scenarios. The architecture consists of a compositional model that divides the arm-hand action recognition problem into three levels. The bottom level is based on a simple but sufficiently accurate algorithm for the computation of the scene flow. The middle level serves to classify action primitives through descriptors obtained from 3D Histogram of Flow (3D-HOF); we further apply a sparse coding (SC) algorithm to deal with noise. Action Primitives are then modelled and classified by linear Support Vector Machines (SVMs), and we propose an on-line algorithm to cope with the real-time recognition of primitive sequences. The top level system synthesises combinations of primitives by means of a syntactic approach. In summary the main contribution of the paper is an incremental method for 3D armhand behaviour modelling and recognition, fully implemented and tested on the iCub robot, allowing it to learn new actions after a single demonstration.
URLhttp://slipguru.disi.unige.it/Downloads/publications/WsRobotVision_Gori_etal2013.pdf

Theme by Danetsoft and Danang Probo Sayekti inspired by Maksimer