Color invariant density estimation for image segmentation and object tracking

In this paper, we formulate a novel density estimation scheme derived from color invariants for image segmentation and object tracking. The advantage of color invariants is that they are robust against varying illumination. However, color invariants are ill-defined when the intensity or saturation is low. Therefore, to achieve robust density estimation, computational methods are presented to estimate the amount of sensor noise through these color invariant images. The obtained uncertainty is subsequently used as a weighting term in the density estimation process to achieve robust image segmentation and object tracking. Experiments are conducted on image sequences recorded from complex 3D scenes. From the experimental results it is shown that the proposed method successfully segments and finds objects robust against illumination and noisy data.