Feature tracking and matching for wide-baseline images with closed-loop sequence

Variations in wide-baseline images occur depending on the viewing angle in a three-dimensional reconstruction of oblique photography, leading to instability in interest-point extraction. Traditional methods, however, invalidate wide-baseline images. In this study, a feature-tracking and matching algorithm based on the analysis of closed-loop images is proposed for wide-baseline images. First, the points of interest in each image were extracted using the SuperPoint algorithm. Continuous pairwise matching was then performed using the SuperGlue algorithm. The matching results were used for feature tracking in both the forward and backward directions, and the feature tracking results were combined. Finally, the algorithm filters the points to obtain an optimal matching result. Comparative experiments demonstrated that this method significantly outperforms the existing conventional methods. The proposed method is robust, obtains matching points more uniformly, and performs better than traditional methods for wide-baseline images.