Efficient Illumination Insensitive Object Tracking by Normalized Gradient Matching

We propose a novel tracking algorithm by minimizing the sum-of-squared differences (SSD) between the normalized image gradients of the template image and the input image from the test image sequence. The proposed tracking algorithm is efficient to implement since it is based on the framework of the inverse compositional algorithm, a computationally efficient tracking technique. The experiments show that the proposed tracking algorithm is superior to the intensity-based inverse compositional algorithm in tracking objects under varying illumination conditions.