This paper presents a method for road navigation,which takes local singular value vectors of image matrix as the feature input and a BP neural network as the classifier. At first,the road image is divided into some sub-images. The algebra feature vectors of all sub-images are extracted using Singular Value Decomposition (SVD) method. With the data of characteristic singular values of sub-images,the Local Singular Value Vector of the whole image are combined,and are used as the input of the classifier. The road images are trained and identified using a BP neural network classifier here. In the experiments,three kinds of road images (left side,right side,and straight direction) are used. Among them,20 pieces of each image are used as training samples,and 30 pieces are used for checking. The results show that the image recognition rate is up to 100% for road navigation with the proposed method.