For extracting the road under complex illumination environment, a road region extraction algorithm is proposed based on color invariant and dark primary color segmentation. The illumination invariant image is obtained with orthogonal decomposition method, and the road information such as vanishing point of image can be extracted based on the clustering segmentation of the dark primary color prior road image. The road region is identified by a constructed soft voting function, and the final accurate road is obtained after morphological processing. In order to effectively evaluate the accuracy and efficiency of the road region extraction algorithm on complex illumination scenes, we construct a complex illumination road scene image dataset for road region extraction and builds benchmark for nine road area extraction algorithms. Experimental results show that the proposed method can effectively reduce the influence of other interference factors such as shadow. Compared with the other eight algorithms, the proposed algorithm can achieve similar accuracy or more accurate than the latest algorithms. Moreover, when the extraction of vanishing point position is accurate, the road extraction performance is more complete and accurate than other algorithms. In addition, the dataset and benchmark built in this paper can effectively evaluate the performance of road region extraction algorithms in complex illumination environment.