In this paper we describe a new technique for detecting the front obstacle from the cruising vehicle using stereo images. To calculate the depth from the camera to the obstacle we have to find out the point of coincidence in two images. However, the process takes much time. So, we attempt to eliminate road surface signs in the left image from the right reverse projection image of the road surface to reduce processing time. Next, we decide the square region that contains an extracted object territory. Then the square region is matched to the right image. This technique was applied to the image acquired on the expressway. As a result, the measurement of the distance to the front obstacle becomes possible.