Road mark recognition using HOG-SVM and correlation

Abstract

In this study we present a novel approach for road mark detection and recognition based on the commercial VIAPIX® module. The proposed approach combines two different techniques, an optical one based on correlation and a numerical technique based on the linear SVM (Support Vector Machine) classifier using HOG (Histogram of Gradient) as descriptor. The first step of our proposed approach consists to applying an inverse perspective mapping of the image acquired by the VIAPIX® module. Then, white color segmentation is applied in order to detect all road marks on the road. Next, a classification of the detected objects is performed using the correlation technique. Finally, the linear SVM technique is used for validating the recognized objects.