ABSTRACT:3D spatial information from unmanned aerial vehicles (UAV) images is usually provided in the form of 3D point clouds. For various UAV applications, it is important to generate dense 3D point clouds automatically from over the entire extent of UAV images. In this paper, we aim to apply image matching for generation of local point clouds over a pair or group of images and global optimization to combine local point clouds over the whole region of interest. We tried to apply two types of image matching, an object space-based matching technique and an image space-based matching technique, and to compare the performance of the two techniques. The object space-based matching used here sets a list of candidate height values for a fixed horizontal position in the object space. For each height, its corresponding image point is calculated and similarity is measured by grey-level correlation. The image space-based matching used here is a modified relaxation matching. We devised a global optimization scheme for finding optimal pairs (or groups) to apply image matching, defining local match region in image-or object-space, and merging local point clouds into a global one. For optimal pair selection, tiepoints among images were extracted and stereo coverage network was defined by forming a maximum spanning tree using the tiepoints. From experiments, we confirmed that through image matching and global optimization, 3D point clouds were generated successfully. However, results also revealed some limitations. In case of image-based matching results, we observed some blanks in 3D point clouds. In case of object space-based matching results, we observed more blunders than image-based matching ones and noisy local height variations. We suspect these might be due to inaccurate orientation parameters. The work in this paper is still ongoing. We will further test our approach with more precise orientation parameters

No abstract

Recently, the mapping industry has been focusing on the possibility of large-scale mapping from unmanned aerial vehicles (UAVs) owing to advantages such as easy operation and cost reduction. In order to produce large-scale maps from UAV images, it is important to obtain precise orientation parameters as well as analyzing the sharpness of they themselves measured through image analysis. For this, various techniques have been developed and are included in most of the commercial UAV image processing software. For mapping, it is equally important to select images that can cover a region of interest (ROI) with the fewest possible images. Otherwise, to map the ROI, one may have to handle too many images, and commercial software does not provide information needed to select images, nor does it explicitly explain how to select images for mapping. For these reasons, stereo mapping of UAV images in particular is time consuming and costly. In order to solve these problems, this study proposes a method to select images intelligently. We can select a minimum number of image pairs to cover the ROI with the fewest possible images. We can also select optimal image pairs to cover the ROI with the most accurate stereo pairs. We group images by strips and generate the initial image pairs. We then apply an intelligent scheme to iteratively select optimal image pairs from the start to the end of an image strip. According to the results of the experiment, the number of images selected is greatly reduced by applying the proposed optimal image–composition algorithm. The selected image pairs produce a dense 3D point cloud over the ROI without any holes. For stereoscopic plotting, the selected image pairs were map the ROI successfully on a digital photogrammetric workstation (DPW) and a digital map covering the ROI is generated. The proposed method should contribute to time and cost reductions in UAV mapping.

A digital surface model (DSM) is an important geospatial infrastructure used in various fields. In this paper, we deal with how to improve the quality of DSMs generated from stereo image matching. During stereo image matching, there are outliers due to mismatches, and non-matching regions due to match failure. Such outliers and non-matching regions have to be corrected accurately and efficiently for high-quality DSM generation. This process has been performed by applying a local distribution model, such as inverse distance weight (IDW), or by forming a triangulated irregular network (TIN). However, if the area of non-matching regions is large, it is not trivial to interpolate elevation values using neighboring cells. In this study, we proposed a new DSM interpolation method using a 3D mesh model, which is more robust to outliers and large holes. We compared mesh-based DSM with IDW-based DSM and analyzed the characteristics of each. The accuracy of the mesh-based DSM was a 2.80 m root mean square error (RMSE), while that for the IDW-based DSM was 3.22 m. While the mesh-based DSM successfully removed empty grid cells and outliers, the IDW-based DSM had sharper object boundaries. Because of the nature of surface reconstruction, object boundaries appeared smoother on the mesh-based DSM. We further propose a method of integrating the two DSMs. The integrated DSM maintains the sharpness of object boundaries without significant accuracy degradation. The contribution of this paper is the use of 3D mesh models (which have mainly been used for 3D visualization) for efficient removal of outliers and non-matching regions without a priori knowledge of surface types.

This paper investigates the feasibility of modelling entire image strips, instead of individual scenes, that have been acquired from the same orbital pass through bundle adjustments. The focus differs from previous researches in the following aspects: firstly, in modelling image strips when control points are available only on a small portion of the strips and, secondly, in comparing the performance of two types of physical sensor models with various sets of adjustment parameters. The models tested were based on modified collinear equations and on satellite orbit and attitude. For each model seven adjustment parameter sets were defined and checked against two SPOT 3 strips of 420 km in length. Results showed that the accuracy of orbit modelling is highly dependent on the choice of parameters to be adjusted. Models with high‐order parameters showed exponential error patterns whereas models with low‐order parameters showed linear error patterns. Among the two types of sensor models, those based on satellite orbit and attitude yielded better orbit modelling performance. In particular, the best performance was achieved by the model that adjusts attitude biases only; a nearly constant accuracy of better than 2 pixels was maintained over the whole orbital segment of 420 km. This supports the feasibility of orbit modelling.