For natural human-robot interaction, the location\nand shape of facial features in a real environment\nmust be identified. One robust method to track facial\nfeatures is by using a particle filter and the active\nappearance model. However, the processing speed of\nthis method is too slow for utilization in practice. In\norder to improve the efficiency of the method, we\npropose two ideas: (1) changing the number of particles\nsituationally, and (2) switching the prediction model\ndepending upon the degree of the importance of each\nparticle using a combination strategy and a clustering\nstrategy. Experimental results show that the proposed\nmethod is about four times faster than the conventional\nmethod using a particle filter and the active appearance\nmodel, without any loss of performance.
Loading....