Research On The Catenary Detection Technology Based On The Three-Dimensional Model

The OCS is an important part of electrified railway power supply system, which is closely related to the safe operation of high-speed electrified railway. Therefore, it is necessary that the accuracy requirement of the OCS real-time detection technology. At present, the contact detection method is commonly applied to the actual system. The research of noncontact detection technology is at the initial stage, and is focused on the research of detection technologies based on the two-dimensional image processing. The bad state detection technology based on three-dimensional information has not been found in the existed literatures. The detection technology based on two-dimensional image is easy to cause the dead corner because of camera angles, exposure and other external factors. While the three-dimensional model contains more abundant information and has better robustness to the two-dimensional image. Therefore, the three-dimensional model based on the detection technology is proposed to detect the bad state of OCS in this paper.For the detection method proposed in this paper, getting the three-dimensional model of the component is the basic of the whole detection. The method to obtain the three dimensional point cloud and the basic process of the reconstruction are presented in this paper.The process of point cloud registration is crucial to the efficiency and accuracy of the entire3D reconstruction process. Scale invariant feature transform (SIFT) algorithm is known as the most widely used local feature-based matching algorithm with high performance, but the intensive computation and high vector dimension of eigenvectors for key points can affect the matching speed. To solve this problem, local binary patterns (LBP) eigenvalues in uniform pattern are used to build eigenvectors after SIFT algorithm extracting key points, and the correspondence of two key points in different point clouds is identified using the strategy based on the ratio of the distance between the nearest key points. Then the coarse registration, fine registration and surface reconstruction are completed, and the3D reconstruction of the OCS parts is finally finished. Experimental results show that the proposed algorithm is able to improve the objective matching speed, and speedup the reconstruction processes.In this paper, the anomaly detection is proposed by based on the obtained three-dimensional model. Segmenting and extracting the different part between the normal model and the exception model and visual displaying. First, the three dimensional model of the parts of the OCS in fault status is reconstructed with the method proposed. In this paper, the insulation sheet defection is adopted. Then, the normal three-dimensional model of the corresponding parts is identified from the known model library by the3D shape context descriptors. Next, the two models are registered by the registration method based on the spin image. The differences between normal model and fault model are segmented and extracted by the nearest neighbor search algorithms based on the Kd-tree. Finally, it is displayed visually that the different part.The validity to obtain three dimensional reconstructions of three dimensional models is verified by the case studies. Meanwhile, the feasibility of the catenary anomaly detection based on the three-dimensional model is verified by the case study of extracting the failure part of insulators.