Camera calibration is a crucial aspect in structured light three-dimensional reconstruction systems, and circular calibration boards are widely employed for camera calibration. During the actual calibration process, the circular calibration images are transformed into ellipses due to the perspective effect of the camera. Additionally, the blurring of edge pixels of circular markers affects the precision of camera calibration. Therefore, a camera calibration method based on subpixel edge detection and perspective projection is proposed. Firstly, the Canny operator with adaptive thresholding is employed to extract the edges of the circular features on the calibration board. Sub-pixel refinement is achieved through interpolation based on the relationships among the edge pixel characteristics, leading to the determination of the ellipse's center. Subsequently, the ellipse feature points are projected into an approximately standard circle using perspective transformation. After extracting the coordinates of the circle center, the points are inversely projected back to the original image. Finally, the Zhang's method is employed to calibrate the camera based on the coordinates of the feature points in the actual images.Experimental results demonstrate that the proposed method reduces the reprojection error by 43.8% compared to traditional methods. The root mean square error for the three-dimensional measurements of standard blocks is minimized at 0.1316mm, and the measurement error ranges between 0.1mm and 0.5mm, meeting the requirements for industrial measurements.