This paper describes an image processing method for recognizing white lines on the road ahead to accomplish automatic lane tracking. The behavior of detected points on the white lines can be regarded as multi-channel time-series data. The stochastic behavior of each point correlates with not only its own past behavior but also the behavior of other points around it. Behavior is identified by using a multi-variable auto-regressive (AR) model. This method can estimate the road configuration ahead logically even if the points cannot be detected clearly because a line is discontinuous or is hidden by a forward vehicle.