Using monocular systems in computer vision has many advantages including low cost, high mobility, etc. In this paper we reconstruct 3-D spatial model using single still image so as to fulfill the task of obstacle detection and road segmentation. In our solution, we represent 3- D spatial model as two important parts: spatial lay out and object depth. Firstly, we present an algorithm to annotate objects in the images such as road, buildings, the n we utilize a state-of-the-art depth estimation algorithm to obta in depth map from the image, and an SVM is used to combine spatial layout and object depth so as to get obstac le detection result. Finally, the experimental resu lts show that this method is simple and effective which has high pract ical value