New sub-pixel interpolation functions for accurate real-time stereo-matching algorithms

Most of the stereo-matching algorithms nowadays need high accuracy, especially for objects at large distances. Lots of approaches are able to provide good results at low costs, but at large distances (small disparities) suffer from the so called “pixellocking effect” i.e. an uneven sub-pixel disparity distribution. In order to compensate this effect, developing sub-pixel interpolation functions can be a viable solution. Previous work has shown that each interpolation function should be adapted to each particular stereo-algorithm and the best way to obtain such a function is trough the process of function fitting. We propose several original interpolation functions for achieving highly accurate sub-pixel disparity maps when applied over the cost volume obtained with two high-speed Semi-Global Matching (SGM) algorithms. In this context we define a multi-objective performance metric for function fitting. To this end we have implemented a tool that provides qualitative measurements of the fitted function by the use of generated ground truth synthetic images, with sub-pixel precision. Real and synthetic test cases provide consistent data - the best interpolation functions for the SGM algorithms have been found and they outperform the state of the art functions, thus the proposed tool is validated.