Combining Particle Filter and Active Shape Models for Lip Tracking

In this paper, a new algorithm for lip tracking is proposed, fusing the strength of stochastic tracker and deterministic tracker. Particle filter and Active Shape Models (ASM) are used as deterministic tracker and stochastic tracker respectively. In order not to build complex dynamic model, we use deterministic tracker instead of stochastic tracker to estimate the local deformation part parameters of shape. Furthermore, to generate the initial optimal parameter sets of deterministic search, Resource Limited Artificial Immune System Clustering (RLAIS) is used, which avoids deterministic tracker searching for each sample. Our method improves experimental results considerably compared to particle filter, ASM or earlier hybrid methods.