Multi-genomic curve extraction

We present Multi-Genomic Curve Extraction (MGCE), a robust method to extract curves in noisy datasets and images. Unlike other robust extraction methods, MGCE does not require to choose the global curve model to extract prior to the process. Instead, it identifies the inliers with respect to an underlying set of local models which number and associated data subsets are automatically determined during the run of the algorithm. As MGCE attempts to minimize this number, the robustness of the inlier extraction is reinforced. The method relies on Multi-Genomic Algorithms (MGA) which are an extension of Genetic Algorithms (GA) designed to handle populations of solutions with variable-length chromosomes. Numerical experiments provide insights about the performance of the method and its applicability to road lane border detection.