| Summary: | 碩士 === 國立成功大學 === 測量及空間資訊學系 === 105 === People are paying more attention to the use of Spherical Panorama Images (SPIs) for its main advantage of wide field of view. Providing accurate location and orientation can enhance more metric application using SPIs. While the exterior orientation parameters (EOPs) of image stations are known, the coordinates of interested points can be determined by space intersection of multiple SPIs. In this study, a special platform called portable panoramic image mapping system (PPIMS) is used to obtain SPIs, and applied for photogrammetric mapping. This system equips with eight single lens cameras and one GNSS receiver, capturing surrounding information simultaneously. After system platform calibration, the images captured with PPIMS at the same image station are combined to be a complete SPI, and then used for mapping application instead of using original images. The EOPs of image stations can be calculated by the network adjustment with multiple SPIs.
No matter in solving image station EOPs or space intersection process, conjugate points selection among overlapped images is necessary. Image matching is considered as an approach to obtain conjugate points much more efficient than manual measurement. In this study, an area-based image matching strategy for automatic conjugate point detection and point coordinate determination with multiple SPIs is proposed. The Sum of Normalized Cross-Correlation (SNCC) and Yet Another Reconstruction Dataprogram (YARD) index are used to check the similarity between images. Within the searching range, similarity profile is generated, and where the maximum similarity locates is regard as the object point position. To decrease the influence caused by scale variations and different FOV between images, the concept of matching in the object space is applied, which uses virtual surfaces for matching by adjusting the scale and perspective of original images, to enhance the matching accuracy.
A test field with five SPIs was set for validation. The root mean square difference (RMSD) of five target points in three directions are (±0.006m, ±0.003m, ±0.004m) in space intersection result, which validates the availability of measurement application using PPIMS. In image matching experiment, four cases with different matching indices and match image type were test. The average maximum similarity of four cases are 0.380, 0.574, 0.573, and 0.696. The RMSDs of selected target points decrease from (±0.014m, ±0.033m, ±0.005m) to (±0.009m, ±0.002m, ±0.004m) with SNCC index, and from (±0.008m, ±0.027m, ±0.005m) to (±0.010m, ±0.008m, ±0.005m) with YARD index. The results reveal that better performance may be achieved using object space matching.
This research shows the feasibility of spatial positioning of interested points with PPIMS SPIs in cm level accuracy, the proposed image matching strategy with PPIMS SPIs is applied and validated. The problem of scale variations and different FOV which causes problem in matching with original images can be improved by object space matching.
|
|---|
