Embranchment labelling and application on the rail measurement

Automatic detecting of rail profile based on computer vision usually uses the optical-section method. The light strip with some important geometry information of the rail surface image is achieved. However, when the sun is on the horizon(early morning and dusk), the noise will appear, and it is difficult to achieve and recognize the light strip correctly if the noise is not removed. This paper aims to remove the noise using algorithm of embranchment labeling for the next thinning and adjustment. After the image is binaried using fixed threshold and the connected components labeled, the parameter of each branch and the intersection table which is the base of collecting the labels can be obtained. According to the feature parameter of each connectivity computed, the light strip of rail profile is then extracted, and the noise removed. The algorithm can support parallel computation. A good effect is achieved on the scene testing of rail auto measurement.