Motion Stereo for Navigation of Autonomous Vehicles
Toshifumi Tsukiyama, Thomas S. Huang · Transactions of the Society of Instrument and Control Engineers · 1986
In order to navigate, vehicles need to deduce their own movements or their positions in given environments. This paper presents a method of motion stereo for finding a vehicle's motion from the 2-d image data taken with a camera mounted on the vehicle.In motion stereo methods, point pairs which correspond to the same object points must be found in the two images and the object points must then be located in the 3-d world at every imaging point. Conventional matching methods have dealt with the case where apparent disparities of point pairs corresponding to the same object points in the two images are relatively small. From the view point of application to navigation of vehicles, taking images at long intervals is desirable, because it requires much computation to process the image data of complex scenes. However, then the camera's direction changes by a large angle from one image to the next. This paper deals with motion stereo using a sparse sequence of single TV images. The approach uses vertices of objects observed in a scene for matching the two images. However, to do matching based on structural information such as vertices of objects derived from 2-d images has some ambiguity. Therefore, range information of vertices of objects is used to find candidate pairs for matching. In addition, the matching is done with the guidance of a base line in the 3-d world which can be seen in both images.The candidate vertices for matching are obtained from analyzing the edge maps of input images using a junction type table and range information of the segments which form the vertices. Range information is obtained by inverse perspective transformation. Some results of experiments using real objects are shown.