In order to improve the accuracy and robustness of lane detection, as well as reduce the influence of illumination change and background interference, an improved Hough transform coupling density space clustering algorithm for lane line detection was proposed. Firstly, the lane line model is established, and the lane boundary is decomposed into a series of small line segments, which are represented by the least square method. Secondly, the improved Hough transform was used to detect the small line segments in the image. A noisy density based spatial clustering of applications with noise was introduced to cluster the extracted small segments, filter out the redundancy and noise in the image, and retain the key information of lane boundary. Then, the gradient direction of the edge pixels was used to define the direction of the small line segments, so that the small line segments on the same side of the boundary have the same direction, while the two small line segments on the opposite lane boundary have the opposite direction. Through the direction function of the small line segments, the candidate clusters of the lane segments were obtained. Finally, according to the candidate clusters, the vanishing point was used to fit the final lane line. It was tested in Caltech data set and the actual road, the data shows that compared with the current popular lane line detection algorithm, under the bad factors such as illumination change and background interference, this algorithm presents more ideal accuracy and robustness, which can accurately identify the normal lane line.