A hybrid tracking framework based on kernel correlation filtering and particle filtering

Recently, the visual object tracking based on correlation filtering has achieved great success. However, there are still some problems need to be improved, such as the scale variation of the target, and so on. Particle filtering (PF) is another commonly used tracking technology. The drawback of PF is that a large number of particles is needed. In this paper, we propose a hybrid tracking framework based on a kernel correlation filtering model and a PF model to complement these two techniques. A local sparse coding is acted as the appearance model of the PF model. First, the kernel correlation filter model is used to obtain the preliminary position of the target. On the basis of the preliminary position of the target, the PF model is used to locate the target further and to capture the scale variation of the target. Finally, both qualitative and quantitative analyses on challenging benchmark with 100 sequences prove the effectiveness of our hybrid tracking framework.