Particle Filter-based Gesture Recognition Technology

This paper uses silhouette as character and B-spline to fit the external shape of gestures. The shape space is introduced as arbitrary processing to the B-spline curve control points will produce some curves which are not similar to the template, and with which the process dimension and computational complexity will be reduced a lot.As particle filter is applicable to any non-linear systems which can be expressed by state space model, this paper uses this method to track the hand. The dynamical model is described as second-order ARP(Auto-Regressive Process), whose dynamical matrices are got by training iteratly with Bootstrap theory. Finally, the particle is weighted by the sum distance between canny feature position and estimated position on the nomal line of sampling points in the curve. The experiments show that this method can achieve tracking robustly.On the basis of hand tracking robustly, we expand the state vector by adding a gesture . category tag. The number of particles is averagely distribued according to the number of gestures categories. When the particles have been assigned based on observations model, the sum weighted value of all particles of each gesture is considered as the evaluation criteria, of which the corresponding categoriy with the largest weight is the final result of the experiment. The experiments show that this method is simple, but it can get the correct rate of 98%, that is to say the gesture recognition method this paper shows is effective...