Abstract: An on-road vehicle detection method under complex illumination environments was introduced. The approach uses the features of a vehicle under complex illumination environment and prior knowledge of the vehicle's front shape based on the hypothesis-verification framework. During the stage of hypothesis generation, edges were extracted from the front image of a vehicle and then fit approximately with the front shape of the vehicle. In the hypothesis verification phase, HOG features were used as a descriptor, in combination with the SVM classifier to complete the verification of hypothesis. The experimental results show that the proposed method works well in complex illumination environment, and it has good performance in detecting vehicle targets under complex illumination environment.