Dynamic gesture recognition based on dynamic Bayesian networks
Part of : WSEAS transactions on business and economics ; Vol.4, No.11, 2007, pages 168-173
Issue:
Pages:
168-173
Author:
Abstract:
Techniques for recognizing and matching dynamic human gestures are becoming increasinglyimportant with the CCTV surveillance system. To provide consistent dynamic gesture recognition system,Hierarchical Dynamic Vision System (HDVS) which based on dynamic Bayesian networks (DBNs) is proposedfor automatically identifying human gestures in this paper. DBNs, directed graphical models of stochasticprocess, generalize HMM by representing the hidden and observed state in terms of state variables in whichcan have more complex interdependencies than HMMs systems do. In this paper, hierarchical hidden Markovmodel (HHMM) is used as the underlying topology in the proposed dynamic system to recognize humangestures with motion trajectories in an indoor scene. A hierarchical HMM, represented by DBN, is structuredmulti-level stochastic processes. In the low-level processing, both motion trajectories and motion directionsgenerated from hand part is used as features after watershed segmentation. In the high-level processing,human gestures are automatically recognized form the inference of HHMM-DBNs. In this paper, we focus onthe following aspects of both system modeling and high-level processing: (1) Completed DBNs structure withHHMM, (2) approaches to human gesture recognition.
Subject (LC):
Keywords:
hierarchical dynamic vision system, dynamic Bayesian network
Notes:
Περιέχει γραφήματα, πίνακες και βιβλιογραφία
