Rectangular region covariance descriptor used in traditional and probabilistic covariance tracking can not reflect
object’s non-rigid deformation precisely. Therefore, a covariance descriptor, elliptical region covariance descriptor, is
proposed, which enables efficient fusion of a diversity of features and modalities into a unified feature model of multi-cue
integration for visual tracking. Further combining this descriptor with Kalman particle filter, a robust visual tracking method
is proposed. Experimental results show the proposed methed has robustness to illumination changes, noise disturbance,
partial and fully occlusion, scale variation, rotation and abrupt object movement.