Continuous Motion Recognition Based on CRF
Beiji Zou · Computer Engineering and Science · 2009
The difficulty in recognizing variable transition postures is one reason of the occurrence of fault continuous motion recognition.We present a new method to tackle the problem by combining CRF(Conditional Random Field) with Condensation(Conditional Density Propagation).In the proposed method,Condensation is used to generate a hypothesis about the border of a movement in the continuous motion sequence,then CRF implements the recognition task of the hypothetical movement fragment.The likelihood of a hypothesis can be evaluated by the probabilistic output of CRF.Experiments show that our method has a considerable improvement in stability and the rate of correct identification compared with CRF.