A visual metrological method based on a two-parallel-plane camera model is proposed for the measurement of the sizes of moving steel plates on production lines. This method uses a data driven mode to calculate the world coordinates of any image point's corresponding projective point on the calibration plane. A k Near Neighbour(k-NN) algorithm is presented to generate the distortion free projective image on the calibration plane from the original image and to build a direct connection between the image and the world coordinate system. An optical center localization algorithm is also proposed. The metrological method uses line structured light to realize the thickness measurement. Then, it extracts plate's edge lines from the distortion free projective image by using their parallel or perpendicular properties and calculates the size of the plate from the edge lines' equations in the world coordinate system. Finally, a framework of large scale steel plate size measurement system is given. The above method is a monocular visual metrology. As comparing to other methods, the method is characterized by simple mounting and smaller calibration process. Experiments results show that if a camera with an image resolution of 640×480 is used to measure a standard 80 mm×50 mm×15 mm aluminum cuboid via the method, the thickness error is 0.1 mm, and the length and width errors are both less than 0.2 mm. When the method is applied to real steel plate measurement, the measurement accuracy is higher than manufacture accuracy, which shows that it satisfies the demands of product measurement.