A detection method for intersection points based on gradient-histogram was proposed. The feature descriptor was generated for each image pixel using gradient of neighborhood pixels. The candidate intersection points were taken from all image pixels according to specific constraints. The candidate intersection points were clustered into different categories based on Euclidean distance. The subpixel positions were computed for each intersection points by the weighting method. The proposed method was validated by simulations. The results show that the method can be used to detect the intersection points of crossing angle 65°~90° with the precision of 0.6 pixel and the detection rate exceeds 90%, and the precision and detection rate of proposed method is invariant to image rotating. The method is robust to image noise and can satisfy the need of detecting intersection points on image under normal prospective transformation. It can be applied to the calibration and surface measurement of structure light 3D reconstruction system.