We present the HD Ground Database, a comprehensive database for ground texture based localization. It contains sequences of a variety of textures, obtained using a downward facing camera. In contrast to existing databases of ground images, the HD Ground Database is larger, has a greater variety of textures, and has a higher image resolution with less motion blur. Also, our database enables the first systematic study of how natural changes of the ground that occur over time affect localization performance, and it allows to examine a teach-and-repeat navigation scenario. We use the HD Ground Database to evaluate four state-of-the-art localization approaches for global localization, localization with the approximate pose being known, and relative localization.

Ground texture based localization is a promising approach to achieve high-accuracy positioning of vehicles. We present a self-contained method that can be used for global localization as well as for subsequent local localization updates, i.e. it allows a robot to localize without any knowledge of its current whereabouts, but it can also take advantage of a prior pose estimate to reduce computation time significantly. Our method is based on a novel matching strategy, which we call identity matching, that is based on compact binary feature descriptors. Identity matching treats pairs of features as matches only if their descriptors are identical. While other methods for global localization are faster to compute, our method reaches higher localization success rates, and can switch to local localization after the initial localization.

Localization using ground texture images recorded with a downward-facing camera is a promising approach to achieve reliable high-accuracy vehicle positioning. A common way to accomplish the task is to focus on prominent features of the ground texture such as stones and cracks. Our results indicate that with an approximately known camera pose it is sufficient to use arbitrary ground regions, i.e. extracting features at random positions without significant loss in localization performance. Additionally, we propose a real-time capable CPU-only localization method based on this idea, and suggest possible improvements for further research.

Two methods to overcome the problems with large vector quantization (VQ) codebooks are lattice VQ (LVQ) and product codes. The approach described in this paper takes advantage of both methods by applying residual VQ with LVQ at all stages. Using LVQ in conjunction with entropy coding is strongly motivated by the fact that entropy constrained but structurally unconstrained VQ design leads to more equally sized VQ cells. The entropy code of the first LVQ stage should aim at exploiting the statistical properties of the source. The refinement LVQ stages quantize the residuals. Simulations show that there exist certain scales of the refinement lattices yielding extraordinary performance. We focus on the search of these scales.

A block motion estimation algorithm which is able to deal with varying gray levels along motion trajectories is introduced. Variations of the illuminating sources and shadows or reflections caused by moving objects are taken into account by the underlying scene model. Orthogonal projection is used to estimate appropriate scaling factors for the luminance components of the image sequence frames. An entropy constrained quantizer for the scaling factor estimates is developed. The proposed algorithm and the standard block matching algorithm are compared with respect to rate-distortion performance and computational load.

Two lattice vector quantization methods are compared. The first, classical method uses Dirichlet domains of lattice points as quantization cells and assigns them reconstruction vectors minimizing the distortion. The second method uses the lattice as codebook but modifies the shapes of the quantization cells by searching for each input vector the mapping onto either the nearest lattice points or one of its neighbors, such that an error criterion subject to an entropy constraint is minimized. Both methods use entropy coding for the codevector indices and an entropy constrained global optimization scheme to find the best lattice scale. Examples demonstrate that the rate-distortion performances of both methods are nearly identical. Hence, especially for non-stationary sources the selection of one of these methods should be based on other criteria: while the first method requires the transmission of the new codebook, the decoder for the second method can permanently adapt itself without side information. The drawback is a higher search complexity for encoding.

Most fractal coding schemes employ an iterative decoding algorithm in order to reconstruct the approximation of the original signal from the fractal code. A necessary condition for obtaining a unique solution is the convergence of the reconstruction process. This paper reports on investigations concerning a necessary and sufficient condition for convergence which is based upon the spectral radius of the transformation matrix. For a very general class of fractal transforms a simple calculation of the spectral radius can be performed in order to decide whether the reconstruction converges or not. This allows more freedom in the choice of the encoding parameters resulting in a better and faster reconstruction process. Also the proposed description leads to a more accurate theoretical foundation of fractal coding schemes.>