Abstract: The method based on computer vision for fast detecting the tread surface flats of railway wheelsets is presented. The line structured light is used to scan the tread surface of wheelsets, thus producing the corresponding deformation light stripe at where it scans. To detect the tread surface flats of wheelsets, the light stripe images collected by the charge coupled device(CCD) camera are processed to get the outline of the tread surface of wheelsets. Based on the characteristics of the light stripe, a new method based on physical model for extracting the center line of light stripe is proposed. Using the energy optimization method, the sub-pixel B-spline center of light stripe can be found when energy is minimized, while greatly reducing the volume of data storage. The proposed approach extracts strips central points at subpixel accuracy successfully and restrains noise, broken lines and divarication efficiently. The center line of light stripe is with a single connectivity. With the rotation of wheelsets, the three-dimensional surface model can be obtained through camera model, thereby realizing non-contact detection of the tread surface flats of wheelsets.

Key words: B-spline, Flats, Physical model, Tread surface of wheelsets