Obstacle detection plays an important role in intelligent vehicle system research. In this paper, the general theory of obstacle detection algorithm is presented based on Ground Plane Transformation. A parametric model for obstacle detection applications is constructed and the influences of the internal, external and pose parameters of the camera on the transformation matrix are discussed. In order to fulfill the requirement in the vibration situations, a new obstacle detection algorithm is proposed based on stereo vision disparity analysis. Experiments show that low computation amount and high reliability can be achieved with the new algorithm.