With the rapid growth of the number of cars,traffic safety has become a hot issue of global concern.According to statistics,the accident caused by lane departure i s increasing year by year.This paper designs a new lane departure warning system based on monocu lar vision.The system processes dynamic image in real time,and alarms when the vehicle deviat es from the lane. Firstly,the paper studies image preprocessing technology,including gray proces sing, intercepting effective region,denoising,image gray enhancement,edge detection and edge repairing.Then,in order to detect the lane whose characteristics are not ap parent or with no lane characteristics,an improved Hough transform and a dynamic detection method are put forward.Next, the paper proposes another algorithm for lane line correction,which is inverse p erspective transformation by four point calibration,turning the lane image into a top view.The system establishes the relation between different coordinate systems,works out the off set of the vehicle, judges the situation and gives directions to the driver.Finaly,a lot of experime nts about the key technology and the whole system are conducted in the real-roa ds.The experimental results show that this design can work well in all kinds of environment and has strong real- time capability and robustness.