Head tracking based on the integration of two different particle filters

Abstract

Existing methods of improving particle filters mainly focus on two aspects: designing a good proposal distribution before sampling and allocating particles to a high posterior area after sampling. An auxiliary particle filter (APF) is one such simple algorithm belonging to the former aspect, which generates particles from an importance distribution depending on a more recent observation. Its weakness is that it requires a large number of particles. On the other hand, a kernel-based particle filter (KPF), which belongs to the latter aspect, is able to greatly reduce the number of particles required and is still able to capture good characteristics of the posterior density. However, a KPF does not take the current observation into account. To utilize their respective strengths, a new algorithm is proposed in this paper with the combination of an APF and a KPF, the APF for designing good proposal density and the KPF for exploring the dominant mode of the posterior density. Experimental results in several real-tracking scenarios demonstrate that the integrated algorithm surpasses the standard particle filter (SPF) when encountering weak dynamic models. Moreover, the proposed algorithm is also able to achieve a comparable performance with KPF whilst reducing computational cost.