Implementation of an Image Sequence Matching Algorithm for Trackside Detection System of Metro Vehicles
Year: 2025
Type: article
Abstract: Relying on the trackside detection system of metro vehicles on Ningbo Line 5, this paper proposes a sequence matching algorithm for the accurate alignment of sample images with reference images, to address issues such as image noise and warping caused by train jolting and environmental changes during detection. The algorithm expands the range of image features through multi-frame stitching and introduces the structural similarity index measure (SSIM) to initially locate the train head. Furthermore, it combines the oriented FAST and rotated BRIEF (ORB) feature point detection algorithm, the K-nearest neighbors (KNN) algorithm, and the random sample consensus (RANSAC) algorithm to ensure accurate matching of feature points and elimination of mismatched points. Quantitative analysis of results from multiple on-site experiments shows that the proposed algorithm can effectively eliminate mismatched points. Under varying conditions such as illumination changes, color differences, and noise interference, it maintained an average re-projection error of less than 1.5 pixels and a correct matching rate of over 92%. The significantly improved robustness and stability of detection system meet the requirements of on-site detection.
Language: Chinese
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7125669073
