Multiplatform Bundle Adjustment Method Supported by Object Structural Information

The registration and integration of data from different platforms are becoming more and more important for real-scene three-dimensional (3-D) reconstruction. In urban areas, the integration of unmanned aerial vehicle images and terrestrial images can compensate for geometric distortions and texture blurring in models generated from single-platform images. However, it remains a question of how to maintain a high accuracy while accounting for the discrepancies between various platforms. Bundle adjustment is a crucial step in building a detailed 3-D model. However, traditional bundle adjustment is usually applied to a single platform. In the case of multiplatform data with significant differences in resolution, flight height, or viewing angles, it can lead to the issues of instability and low accuracy in solving bundle adjustment problems. This article innovatively proposes a multiplatform bundle adjustment method, which is supported by object structural information. First, the method performs patch-based matching of images from different platforms and obtains cross-platform tie points. Second, refined patches obtain object structural information by calculating the depth values and ground sampling distances of image points. Finally, multiplatform bundle adjustment is conducted using weights calculated for both object and image points based on factors obtained in the second step. The experimental results show that, in general, the proposed method can achieve the accuracy level required for practical applications. Compared with the bundle adjustment method without object structural information, the improvement of accuracy is significant, with an average improvement of 53.38% across the four datasets.