To solve the road oriented marking recognition problem in complex driving environment

this paper presents a novel approach with two steps: 1) merge multi scale autoconvolution (MSA) moment with affine invariant and image region property to define a new feature descriptor to describe road oriented marking; 2) apply support vector machine (SVM) to classify the image objects. The experimental data demonstrate that the new descriptor has an approximate projective invariance and is suitable for road oriented marking classification in different visual angles and partly being occluded .