Towards real-time image segmentation using polynomial functions

We present current developments towards a real-time method for segmentation of images for robotics applications. Our algorithm follows a fast scan line procedure (rows and columns), thus breaking the 2D segmentation problem in two 1D processes. In the first phase we use polynomial functions to find possible 1D boundaries in the image columns and rows separately. In a second step, a fusion method is applied, removing redundancies and false boundaries to find the regions of interest in the segmented images. We try to improve the algorithm using a coarse to fine multi-resolution approach. Preliminary results demonstrate possible applicability in real-time.