Abstract: In order to ensure the safety of metro train running, stopping and passenger landing, it is necessary to check the metro platform gauge and line maintenance periodically during long-term operation. At present, the regular contact detection has many disadvantages, such as low detection efficiency and greater human influencing factors. A portable non-contact platform gauge detector based on the machine vision technology is developed. Then, a calibration method based on reference point matching is put forward by adopting the triangulation principle, aiming to solve the problem of large-scale photogrammetry from the close-range oblique angle. Through field test and data analysis, the measurement accuracy could meet the detection requirements of metro station platform boundary intrusion.