Feature Point-Based Dense Image Matching Algorithm for 3-D Capture in Terrestrial Applications

Abstract

Background and Objective: Image matching is a bottleneck that must be surpassed in photogrammetric measurement, camera calibration and computer vision. This study investigated the performances of the well-known feature point detectors in automatic matching. Materials and Methods: A large base-to-height ratio of stereo images creates perspective distortion and the selection of an objectʼs small shape properties from the image becomes difficult. Therefore, a large base-to-height ratio affects matching and measurement accuracy in photogrammetry. The relative variations on the scale of stereo images also make it difficult to create conjugate points between them. Different base-to-height and various scale stereo images were evaluated to compare the matching performance of these operators. Results: The results show that the number of matched feature points decreases when the base-to-height ratio of the images is increased. The SIFT, ASIFT and SURF operators did not match the images with a base-to-height ratio larger than 1.5 and a scale change of more than three times. Finally, ASIFT generated more matched points than SIFT and SURF. Conclusion: These findings are useful for automatic three-dimensional measurement from stereo images. Three-dimensional measurement can be performed with fewer images without any force on computer capacity. The imaging positions and point of view angles for multiview evaluations should also be planned according to the limitations of these operators.