In this paper, we proposed a new tracking algorithm that
can simultaneously overcome the difficulties associated with
drastic illumination change, partial occlusion, a similar colored
background, and low illumination. For the proposed tracking
algorithm, we introduced the binary pattern-based SBP model
that consists of a set of multiple SBPs. In addition, we
proposed a kernel-based similarity measure between two SBP
models for target localization. To further improve the tracking
performance, we also employed the CAT method along with
the SBP-based tracking method.
Since the proposed SBP model is invariant with respect
to the gray-scale change, our tracking algorithm can tolerate
abrupt illumination variations, which often occur in natural
scenes. The binary pattern-based SBP model also provides
better discrimination capabilities in the situation of similar
colored region or low illumination where the color-based
model tends to fail to track the target. Furthermore, the SBP
model can cope successfully with the problem of partial
occlusion by containing the information of spatial relationship
between subregions within the target object region. Both
qualitative and quantitative analysis results show that the
proposed algorithm outperforms the existing algorithms in the
previously mentioned tracking environment.
As a future work, we plan to develop an algorithm that
detects the tracking failure and recovers the tracking process
when the target tracking fails due to long duration of heavy
occlusion.