A model-based complex background gesture recognition system

In this paper, a hand gesture recognition system is introduced to recognize the gestures of sequential digital number in a complex background. We present a method which uses the flexible models to describe the hand shapes and its variability. First, we present the hand shapes in the training set by the manually labeled points located on the boundary of the objects. Next, we align the all training shapes for examining the statistics of the coordinates of the model points over the training set. This flexible model can only fit the new hand examples which are similar to the shapes of the corresponding training set. From the extracted hand features, we can judge which state it belong to, and present the entire continuous gesture by a state sequence. In the experiments, we illustrate that the system can recognize the gestures of both single digital number and sequential digital number