Authors have been studying a guidance and control system for an autonomus underwater vehicle of a streamlined cruising type, which is appropriate for deep sea investigation in a wide area for its high efficiency of energy consumption, and it does not need safety equipment for human operators. However, in the unmanned control, real time utilization of visual information for control and other decision making, which is a major advantage of human operation, is difficult at present. In the study, basic methods to estimate motion or attitude of the vehicle (camera) from video data which is obtained continuously by a video recorder and/or a camera are investigated, and several alternative methods to estimate the vehicle motion are evaluated by actual sample pictures. Finally, possibility of the motion estimation from visual information is investigated by applying those methods to actual underwater video images. From the study, it is assured that estimation of camera motion (ego motion) from usual landscape pictures is possible by simpler and quicker methods. on the other hand, it is not so easy at the sea bottom, for there are few landmarks and visual pattern does not change clearly. However, by using full information and selecting appropriate estimating points, possibility of the motion estimation at the sea bottom is assured for the case human operator can recognize the motion visually.