Abstract
Mobile Mapping’s ability to acquire high-resolution ground data is opposing unreliable localisation capabilities of satellite-based positioning systems in urban areas. Buildings shape canyons impeding a direct line-of-sight to navigation satellites resulting in a deficiency to accurately estimate the mobile platform’s position. Consequently, acquired data products’ positioning quality is considerably diminished. This issue has been widely addressed in the literature and research projects. However, a consistent compliance of sub-decimetre accuracy as well as a correction of errors in height remain unsolved. <br><br> We propose a novel approach to enhance Mobile Mapping (MM) image orientation based on the utilisation of highly accurate orientation parameters derived from aerial imagery. In addition to that, the diminished exterior orientation parameters of the MM platform will be utilised as they enable the application of accurate matching techniques needed to derive reliable tie information. This tie information will then be used within an adjustment solution to correct affected MM data. <br><br> This paper presents an advanced feature matching procedure as a prerequisite to the aforementioned orientation update. MM data is ortho-projected to gain a higher resemblance to aerial nadir data simplifying the images’ geometry for matching. By utilising MM exterior orientation parameters, search windows may be used in conjunction with a selective keypoint detection and template matching. Originating from different sensor systems, however, difficulties arise with respect to changes in illumination, radiometry and a different original perspective. To respond to these challenges for feature detection, the procedure relies on detecting keypoints in only one image. <br><br> Initial tests indicate a considerable improvement in comparison to classic detector/descriptor approaches in this particular matching scenario. This method leads to a significant reduction of outliers due to the limited availability of putative matches and the utilisation of templates instead of feature descriptors. In our experiments discussed in this paper, typical urban scenes have been used for evaluating the proposed method. Even though no additional outlier removal techniques have been used, our method yields almost 90% of correct correspondences. However, repetitive image patterns may still induce ambiguities which cannot be fully averted by this technique. Hence and besides, possible advancements will be briefly presented.
Highlights
Mobile Mapping data products are a valuable, additional source of geo-information especially to extend coverage and enhance detail in urban areas
Many authors rely on other sources of exterior orientation information, such as digital maps, aerial imagery or ground control points (Ji, Shi et al (2015); Jaud, Rouveure et al (2013); Kümmerle, Steder et al (2011); Levinson and Thrun (2007))
This paper will focus solely on the registration between Mobile Mapping (MM) and ortho-images computed from aerial nadir images, it constitutes the basis for further developments with respect to the registration of MM and oblique images
Summary
Mobile Mapping data products are a valuable, additional source of geo-information especially to extend coverage and enhance detail in urban areas. The majority of these methods utilises mobile laser scanning data, and approaches relying on MM images, such as Ji, Shi et al (2015) compensate for matching errors within the filtering stage rendering a reliable registration unnecessary by accepting correct but mediocre correspondences These methods do not compensate for vertical errors, and cannot comply with a consistent decimetre accuracy. Stateof-the-art algorithms for the extraction of feature keypoints can account for differences in scale, rotation, illumination and perspective all to a certain degree (Alcanterilla, Nuevo et al (2013); Rublee, Rabaud et al (2011); Levi and Hassner (2015)), but bridging great overall variations e.g. between MM and aerial images has not been achieved yet. Depending on the distribution and the number of correspondences, they may either serve as an outlier mask for a subsequent feature matching or already as an input for an adjustment to rectify the MM data set
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
