Chi-square Fitting of Deformable Templates for Lane Boundary Detection

An accelerated algorithm for lane boundary detection is described. It combines random search with the chi-square fitting to obtain the best set of parameters of a deformable template. The lane boundaries found in a grey level image are modelled via deformable templates. To find the best fitting template a likelihood function is used. It is based on the degree of match (in magnitude and direction) between the deformable template and the underlying lane boundaries. Chi–square fitting is applied on a randomly chosen template to maximise the likelihood function. The experimental results demonstrate the performance of this improved algorithm.