Résumé : |
(Auteur) This dissertation addresses the topic of automatic Digital Surface Model (DSM) generation from linear array images. Research on this issue is mainly motivated by the following facts: In recent years, CCD linear array sensors are widely used to acquire high-quality, high-resolution panchromatic and multispectral imagery for photogrammetric and remote sensing applications. Most of these sensors have the ability to acquire more than 2 views of the terrain surface during a single flight line or orbit. The processing of these kinds of images provides a challenge for algorithmic redesign and this opens the possibility to reconsider and improve many photogrammetric processing components. In particular, automatic DSM generation through image matching techniques is one of the main topics. Although this topic has gained much attention in the past years and a wide variety of approaches have been developed, the accuracy performance and the problems encountered are very similar in the major approaches and the performance of these approaches does by far not live up to the standards set by manual measurements. Therefore, efforts have to be made to develop a general framework for automatic DSM generation from linear array images, into which specific algorithms can be inserted easily, investigated and combined in order to achieve reasonable results in terms of precision and reliability.
In this dissertation, an image matching approach for automatic DSM generation from linear array images, which has the ability to provide dense, precise and reliable results is presented. The approach integrates different matching primitives, uses available and explicit knowledge concerning the image geometry and radiometry information, combines several image matching algorithms and automatic quality control, and works with a coarse-to-fine hierarchical matching strategy. The most outstanding characteristics are the efficient utilisation of multiple images and the integration of multiple matching primitives. With this approach, the linear array images and the given or previously triangulated orientation elements are taken as inputs. After pre-processing of the original images and production of the image pyramids, the matches of 3 kinds of features, i.e. feature points, grid points and edges, on the original resolution images are finally found progressively starting from the low density features on the images with the lowest resolution. An intermediate DSM is reconstructed from the matched features on each level of the pyramid by using the constrained Delauney triangulation method, which in turn is used in the subsequent pyramid level for the approximations and self-tuning of the matching parameters. Finally least squares matching methods are used to achieve more precise matches for al] the matched features and identify some false matches.
The presented approach essentially consists of several mutually connected components: the image pre-processing, the multiple primitive multi-image (MPM) matching, the refined matching and the system performance evaluation. Each of them is important and possesses particular features, which are fully elaborated in different parts of the dissertation.
First of all, a pre-processing method, which combines an adaptive smoothing filter and the Wallis filter, is used in order to reduce the effects of the inherent radiometric problems and optimize the images for subsequent feature extraction and image matching procedure. The method mainly consists of 3 processing stages. In the first stage, the noise characteristics of the images are analyzed quantitatively in both homogeneous and non-homogeneous image regions. The image blur problem (image unsharpness) is also addressed through the analysis of the image's Modulation Transfer Function (MTF). Then, an adaptive smoothing filter is applied to reduce the noise level and at the same time, to sharpen edges and preserve even fine detail such as corners and line end-points. Finally, the Wallis filter is applied to strongly enhance and sharpen the already existing texture patterns.
The MPM matching procedure is the core of our approach. In this approach, the matching is performed with the aid of multiple images, incorporating multiple matching primitives feature points, grid points and edges, integrating local and global image information and, utilizing a coarse-to-fine hierarchical matching strategy.
The MPM approach mainly consists of 3 integrated subsystems: point extraction and matching procedure, edge extraction and matching procedure and relational matching procedure. These 3 subsystems are followed through the image pyramid and the results at higher levels are used for guidance at the lower levels. At each pyramid level, the correspondence is established in two matching stages - locally and globally. In the local matching stage dense patterns of points and edges are matched. A unique and robust matching algorithm - The Geometrically Constrained Cross-Correlation (GC3 ) algorithm is employed to provide matching candidates for points and edge pixels. The algorithm. is based on the concept of multi-image matching guided from the object space and allows reconstruction of 3D objects by matching all the images at the same time, without having to go through the processing of all individual stereo-pairs and the merging of all stereo-pair results. The GC3 method, with the self-tuning of the parameters, leads to a reduction of problems caused by occlusions, multiple solutions and surface discontinuities. The global matching stage is responsible for imposing global consistency among the candidate matches in order to disambiguate the multiple candidates and avoid mismatches. The global matching is resolved by a probability relaxation based relational matching method. It uses the local support provided by points within a 2D neighbourhood. This corresponds to imposing a piecewise smoothness constraint, in which the matched edges serve as breaklines in order to prohibit the smoothness constraint crossing these edges and preserves the surface discontinuities.
The modified Multiphoto Geometrically Constrained Matching (MPGC) and the Least Squares B-Spline Snakes (LSB-Snakes) methods are used to achieve potentially sub-pixel accuracy matches and identify some inaccurate and possibly false matches. The DSM derived from the MPM module provides good enough approximations for these methods and increases the convergence rate. The initial values of the shaping parameters in MPGC matching can also be predetermined by using the image geometry and the derived DSM data. Finally, for each matched point, a reliability indicator is assigned based on the analysis of the matching results. For edges, a simplified version of the LSB-Snakes is implemented to match the edges, which are represented by parametric linear B-spline functions in object space. With this method, the parameters of linear B-spline functions of the edges in object space are directly estimated, together with the matching parameters in the image spaces of multiple images.
The system bas been tested extensively of linear array images with different image resolution and over different land cover types. The accuracy evaluation is based on the comparison between high quality DEMs/DSMs derived from airborne Laser Scanner or manual measurements and the automatically extracted DSMs. As evidenced by the visual inspection of the results, we can reproduce not only the general geomorphological features of the terrain relief, but also detailed features of relief. The results from the quantitative accuracy test indicate that the presented concept has the capability to give good and encouraging results. If the bias introduced by trees and buildings is taken out, we can expect a height accuracy of one pixel or even better from satellite imagery as "best case" scenario. In case of very high resolution TLS/SI images (footprint 8 cm and better) it is obvious that the "one pixel rule" cannot be maintained any more. Alone surface roughness and modeling errors will lead to larger deviations, such that an accuracy of 2 to 5 pixels should be considered an acceptable result. |