Steel ball is a key assembly part of bearing, and its surface defect is one of the main causes of bearing failure. At present, the domestic inspection of the surface quality of steel balls is still using the artificial visual inspection method with low detection efficiency and high false detection rate. This paper proposes a detection method for steel ball surface defects based on optical dark field line scanning technology, and constructs a surface defect detection system for high reflectivity steel ball by dark field line scanning.As the surface of the bearing steel ball is close to the mirror reflection, the traditional light source will cause the surface of the halo, resulting in local overexposure of the camera, and the image of the surrounding object will be reflected into the surface of the steel ball, and part of the defect information will be submerged. In addition, because the normal direction of the steel ball changes everywhere, the lighting source used in the traditional detection method is difficult to construct an optical dark field for the sphere. Given the above two cases, this paper uses line laser scanning to construct the spherical optical dark field, and simulates the reflected light direction under different incidence angles and line widths, and finally solves the detection blind area caused by overexposure and environmental image reflection on the surface of the steel ball, and realizes the construction of the broadband dark field area on the surface of the steel ball and the acquisition of images.The spherical geometry of the steel ball makes it impossible for all surface defects to be presented coherently without distortion on the same two-dimensional image. That results in missed or false detection due to incomplete defect shooting. Therefore, this paper stitches multiple spherical and ring images to cover the whole sphere, studies its distortion correction algorithm, filters the image noise through median filtering and threshold segmentation, extracts defect features, establishes the spatial geometric relationship among the points of the image, and develops the three-dimensional reconstruction algorithm of the sphere surface feature points. The surface defects of different spherical images are reconstructed on the same three-dimensional spherical surface. The discrete defect point cloud on the three-dimensional spherical surface is divided into different classes by clustering segmentation. The point cloud class above the set threshold is screened by triangulation algorithm. The three-dimensional continuous detection of surface defects of bearing steel ball is realized, which opens up a new idea for the detection method of surface defects of steel ball.The light and dark field defect detection comparative experiment was carried out on the bearing steel ball with a diameter of 6 mm and a precision of G20. The result shows that the dark field had obvious technical advantages compared with the bright field. The repeatability of the proposed method is 0.14% for non-destructive balls with intact surfaces and 0.11% for wear balls with surface defects. The designed system can realize three-dimensional reconstruction and detection of small defects on the surface of steel balls.