Lane departure warning system is an important element in improving driving safety and the main difficulty of which is to detect and recognize lane markings quickly and precisely. In this paper, for satisfying the quickness and precision requirement of recognizing lane markings, a novel vision-based lane markings detection and recognition method, which mainly comprise of two steps, is proposed. The first step of which is to roughly determine the ROI of lane markings by means of projection, while the next is to precisely detect lane markings by combining Hough transform with the least square method to fit straight-line. Large amounts of experiment indicate that the proposed method can rapidly and precisely recognize lane markings from sequential images.