Abstract: Sensors are widely used in railway safety maintenance. In order to achieve real-time online gauge measurement, an on-board machine vision system of gauge detection was established with the combination of laser technique and CCD image sensors. Firstly, the working principle and structure of the system were introduced. A fast extraction method of rail contour centerline was proposed based on pre-noising and distance transform (PNDT) algorithm. Then, the image was enhanced with the strong contrast stretching method and exponential transform method. Fast pre-noising processing was conducted with the combination of Gaussian smoothing and dynamic region of interest (ROI). Finally, precise threshold segmentation processing was carried out for the images. The rail contour centerline was gotten and the gauge measurement point was located with the distance transform method. Experimental results show that the detection accuracy of the system is -1mm~+1mm and image frame processing speed is 14.35ms. The system can meet the requirement of the robustness, detection speed and accuracy in real-time online gauge detection.