Research on railway track edge detection based on BM3D and Zernike moments

With the rapid development of intelligent rail transportation, the realization of intelligent detection of railroad foreign body intrusion has become an important topic of current research. Accurate detection of rail edge location, and then delineate the danger area is the premise and basis for railroad track foreign object intrusion detection. The application of a single edge detection algorithm in the process of rail identification is likely to cause the problem of missing important edges and weak gradient change edges of railroad tracks. It will affect the subsequent detection of track foreign objects. A combined global and local edge detection method is proposed to detect the edges of railroad tracks. In the global pixel-level edge detection, an improved blok-matching and 3D filtering (BM3D) algorithm combined with bilateral filtering is used for denoising to eliminate the interference information in the complex environment. Then the gradient direction is added to the Canny operator, the computational template is increased to achieve non-extreme value suppression, and the Otsu thresholding segmentation algorithm is used for thresholding improvement. It can effectively suppress noise while preserving image details, and improve the accuracy and efficiency of detection at the pixel level. For local subpixel-level edge detection, the improved Zernike moment algorithm is used to extract the edges of the obtained pixel-level images and obtain the corresponding subpixel-level images. It can enhance the extraction of tiny feature edges, effectively reduce the computational effort and obtain the subpixel edges of the orbit images. The experimental results show that compared with other improved algorithms, the method proposed in this paper can effectively extract the track edges of the detected images with higher accuracy, better preserve the track edge features, reduce the appearance of pseudo-edges, and shorten the edge detection time with certain noise immunity, which provides a reliable basis for subsequent track detection and analysis.