Abstract: Vision based road detection is a key technique in autonomous driving and advance drive assistance system. An approach for vision-based road detection which exploits road shape prior and online discrimination analysis was proposed. With available annotated road images, road shape dictionary was constructed, and sparse representation was applied to classify specific shape of input road images, the road prior which characterizes the layout of the road and off-road region was obtained. Then an online discrimination analysis was applied to choose the best discriminative color channel in which road detection was realized by simple image segmentation. Experimental results on both public and our own database demonstrate the effectiveness of the proposed road detection method.