Towards a Unified Approach to Homography Estimation Using Image Features and Pixel Intensities

The homography matrix is a key component in various vision-based robotic\ntasks. Traditionally, homography estimation algorithms are classified into\nfeature- or intensity-based. The main advantages of the latter are their\nversatility, accuracy, and robustness to arbitrary illumination changes. On the\nother hand, they have a smaller domain of convergence than the feature-based\nsolutions. Their combination is hence promising, but existing techniques only\napply them sequentially. This paper proposes a new hybrid method that unifies\nboth classes into a single nonlinear optimization procedure, applies the same\nminimization method, and uses the same homography parametrization and warping\nfunction. Experimental validation using a classical testing framework shows\nthat the proposed unified approach has improved convergence properties compared\nto each individual class. These are also demonstrated in a visual tracking\napplication. As a final contribution, our ready-to-use implementation of the\nalgorithm is made publicly available to the research community.\n