The problem of judging direction of the road is a tough problem in navigation system. This paper presents a image-based approach to address the direction problem in the road using Support Vector Machine(SVM) and Bayesian rules. Established localization or orientation algorithm is infeasible in the direction problem due to its time and space complexity. SVM is employed as an effective model for classification and Bayesian rule can improve the accuracy integrating more than one images. Experimental result shows that the proposed system reaches high accuracy in the direction problem.