Charging stations serve as essential power infrastructure for unmanned vehicles, where precise localization enables reliable operation. Current positioning algorithms predominantly use template matching and deep learning, yet both face limitations. Template matching performs poorly under perspective changes, while deep learning lacks real-time applicability. To address these issues, this article proposes an improved ORB feature matching algorithm incorporating deblurring and color-invariant processing for scale-invariant charging station localization. The method first applies a multi-scale pyramid and fuzzy layer segmentation for image deblurring. Next, a color invariant model preprocesses template and test images to extract invariant features. A scale space is then constructed for these features, with Fast-Hessian detecting extremum points to obtain scale-invariant keypoints. Feature descriptors are computed using rBRIEF, while Hamming distance and an accelerated RANSAC filter mismatches to derive the inter-image mapping matrix. The charging station′s pose is finally estimated using cooperative target dimensions and PnP. Experimental results show superior deblurring performance over traditional methods. Compared with the conventional ORB, the proposed algorithm resolves feature extraction failures in the same-gray but different-color regions, enhances matching accuracy, ensures scale invariance, and improves feature distribution uniformity, ultimately boosting positioning precision.