Moving object tracking under varying illumination conditions

In this paper, a novel approach is proposed for tracking non-rigid moving objects with a stationary camera under varying illumination conditions. By using the well-known Bayesian framework, our method combines a specially designed color model and a texture model through a level set partial differential function. Different from traditional methods, our color and texture models can extract robust information that is insensitive to illumination variations. This makes it feasible to determine whether temporal variations in images are caused by object motion or illumination changes. Moving objects can then be tracked robustly and accurately in spite of abrupt illumination variations. Since no prior shape information about moving objects is required, it is especially suitable for the situation where shape information is hard to be obtained. Experiments show that this method has a great capability to track non-rigid moving objects under globally or locally varying illumination conditions, even when light intensities change abruptly.