Road Lane Tracking Using Constrained Scale Space Radon Transform

On a road, there are often parallel linear structures painted in white or color to demarcate the lanes. For automatic driving, these structures must be tracked in image sequences acquired in real time by an embedded camera so that a vehicle can stay in its lane or change lanes. These road markings appear in grayscale images as continuous or discontinuous linear structures of variable width, converging at vanishing points. In this article, we propose tracking these road markings using both the Scale Space Radon Transform (SSRT) and projective geometry. These markings are considered as a mixture of linear structures, with the model being updated under parallelism constraints. The proposed multiline tracker is less sensitive to noise and acquisition conditions due to simultaneous adjustment expressed in feature space. Experimental results demonstrate the superiority of SSRT and the simultaneous tracking of multiple lines.