In the process of rail surface defect detection, the collected image data is the original data, which is the basis of the whole system. Their quality determines the accuracy of later defect extraction and the speed of the algorithm. In order to collect high-quality rail surface images, this paper proposes an optical theoretical model for rail surface defect detection that can provide stable and uniform illumination conditions, analyses the relationship between image histogram and image quality, and studies a reflective surface that can avoid. The light source directly illuminates the surface of the rail. By designing a reasonable lighting system, 270-degree stable and uniform lighting is provided for the rail surface. Experiments demonstrate the rationality of the proposed model. The results show that the stable uniform illumination enhances the contrast of defects and background, which facilitates image recognition in post-image processing. Moreover, the incident angle of the uniform illumination is evenly distributed, which effectively eliminates the image quality influence caused by vibration and other reasons during the detection process.