Rail Profile Detection Based on PNP Algorithm

With the development of domestic high-speed railway industry, the increased frequency of rail use makes the rail wear rate higher, and the service life of rail also becomes shorter. Excessive wear of rail will pose a great threat to the safety of the train, which makes it necessary for the work of regular rail profile detection and grinding. Nowadays, the main rail measurement method is mechanical contact measurement. this method can only measure the geometric parameters of the rail profile qualitatively. The shortcomings of this method are that the rail can not be measured in all directions, and it requires manual operation and has low efficiency. In this paper, we propose a rail profile detection method based on structured light, which can extract the rail profile flexibly and accurately. The key step in this method is to calculate the camera pose according to PNP algorithm. PNP algorithm used in this system is to known 3D-2D point pairs to restore the camera posture accurately. However, because of the great interference in the calculation of the camera pose caused by wrong 2D-3D points, the existing PNP algorithm is sensitive to noise and less robust. In this case, we add random sample consensus algorithm to process the input points to filter out the wrong pairs of points. At last, the error model of PNP camera pose is constructed by Levenberg-Marquart algorithm and the least square optimization is carried out, which makes the system more stable, more robust and more accurate.