The problem of autonomous driving and co-navigating has received increasing attention in recent years. This paper researches on road recognition and tracking in automatic visual road vehicle guidance system by processing and analysing a long sequence stereo images of real road which are gathered on an autonomous vehicle driveing outside. An algorithm is proposed for estimating the parameters of three-dimensional motion

based on long sequence stereo images processing. It is our core method to build the road's curvature dynamic model

vehicle's state kinematic model and camera's perspective model

and to estimate the 3D motion parameters using Kalman filter. The algorithm is fast

robust and practical. Experimental results with real scene images are given. Experiments prove that the results are satisfying for road recognition and tracking

so the approach could actively improve the research on the performance of the visual road vehicle guidance system.