This letter presents a novel approach to automated extraction of roof planes from airborne light detection and ranging data based on spectral clustering of straight-line segments. The straight-line segments are derived from laser scan lines, and 3-D line geometry analysis is employed to identify coplanar line segments so as to avoid skew lines in plane estimation. Spectral analysis reveals the spectrum of the adjacency matrix formed by the straight-line segments. Spectral clustering is then performed in feature space where the clusters are more prominent, resulting in a more robust extraction of roof planes. The proposed approach has been tested on ISPRS benchmark data sets, with the results showing high quality in terms of completeness, correctness, and geometrical accuracy, thus confirming that the proposed approach can extract roof planes both accurately and efficiently.

Automatic registration of multisensor data, for example, imagery and Light Detection And Ranging (LiDAR), is a basic step in data fusion in the field of geospatial information processing. Mutual information (MI) has recently attracted research attention as a statistical similarity measure for intensity-based registration of multisensor images in the related fields of computer vision and remote sensing. Since MI-based registration methods rely on joint probability density functions (pdfs) for the data sets, errors in pdf estimation can affect the MI value, causing registration failure due to the presence of nonmonotonic surfaces of similarity measure. The quality of the estimated pdf is highly dependent upon both the bin size and the smoothing technique used in the pdf estimation procedure. The lack of a general approach to assign an appropriate bin size value for the pdf of multisensor data reduces both the level of automation and the robustness of the registration. In this paper, a novel bin size selection approach is proposed to improve registration reliability. The proposed method determines the best (uniform or variable) bin size for the pdf estimation via an analysis of the relationship between the similarity measure values of the data and the adopted geometric transformation. This highlights the role of the component of MI sensitive to the transformation, rather than the MI component that is unrelated to the transformation, such as noise. The performance of the proposed method for the registration of aerial imagery to LiDAR point clouds is investigated, and experimental results are compared with those achieved through a feature-based registration method.

Geometric and radiometric attributes of targets are provided by full-waveform LiDAR data. However, the accuracy of such information depends largely on the adopted data processing method. In this study, the emphasis is on the retrieval of the temporal target cross-section by regularization methods, with the subsequent extraction of the backscattering cross-section (BCS) and backscatter coefficient (BC), the aim being to characterize different classes in an urban scene. In particular, a sparsity constraint regularization method has been investigated to provide a temporal target response with high resolution. The L-curve method is represented as a proper approach for estimation of the optimal regularization parameter, where a polynomial function is fitted to a group of discrete points associated with the corresponding values between the two terms in the objective function. The proposed methods have been tested with real full-waveform LiDAR data, demonstrating the capability of efficient separation of targets in the waveform signal.

Accurate waveform restoration, from the received noisy waveform, is of great interest to the full-waveform LiDAR community. As a result of this, important attributes could be estimated precisely which are valuable in describing and differentiating LiDAR targets. Assumptions behind prominent methods like the Gaussian decomposition do not hold due to the complexity of the land surface. Deconvolution is a standard approach to retrieve the target cross-section. A regularization method is proposed based on sparsity constraints and it is compared to other well-known deconvolution methods. Numerical and visual results illustrate the robustness of the proposed method with regard to signal restoration and to suppression of noise and oscillation effects.

Automatic registration of multi-sensor data is a basic step in data fusion applications. Mutual information (MI) has been widely used in medical and remote sensing image registration. In this paper, an effective histogram binning technique is proposed to improve the robustness of image registration using MI and Normalized MI (NMI). Increasing the bin size improves the robustness of MI to local maxima that occur in the convergence surface of MI. In addition, the computation cost of registration is decreased due to use of a smaller joint pdf, without decreasing the accuracy. The performance of the proposed method in the registration of aerial imagery with LiDAR data has been experimentally evaluated and the results obtained are presented.

Automatic 3D extraction of building roofs from remotely sensed data is important for many applications including city modeling. This paper proposes a new method for automatic 3D roof extraction through an effective integration of LIDAR (Light Detection And Ranging) data and multispectral orthoimagery. Using the ground height from a DEM (Digital Elevation Model), the raw LIDAR points are separated into two groups. The first group contains the ground points that are exploited to constitute a ‘ground mask’. The second group contains the non-ground points which are segmented using an innovative image line guided segmentation technique to extract the roof planes. The image lines extracted from the grey-scale version of the orthoimage are classified into several classes such as ‘ground’, ‘tree’, ‘roof edge’ and ‘roof ridge’ using the ground mask and colour and texture information from the orthoimagery. During roof plane extraction the lines from the later two classes are used to fit roof planes to the neighbouring non-ground LIDAR points. Finally, a new rule-based procedure is applied to remove planes constructed on trees. Experimental results show that the proposed method successfully removes vegetation and offers high extraction rates.

Separation of buildings from trees is a major challenge in automatic building detection. In residential and hilly areas, buildings are often surrounded by dense vegetation. This paper presents a three-step method for effective separation of buildings from trees. Firstly, height and width thresholds are applied to LIDAR data for removing small bushes and trees with small horizontal coverage, respectively. The generation of the building mask, where each black region indicates a void area from which there are no laser returns below the height threshold, also helps in separation of buildings from the nearby trees. Then image entropy and colour information are applied together to remove trees exhibiting high texture. Finally, an innovative rule-based procedure is employed using the edge orientation histogram from the imagery to eliminate the remaining trees. Experimental results show that the algorithm offers high building detection rate in complex scenes which are hilly and densely vegetated.

Although a large number of building roof reconstruction techniques from the remotely sensed data have been proposed, there is a serious lack in standard and automatic evaluation system. The manual evaluation systems are based on human judgment and so may be biased. Moreover, the application of such an evaluation system on a large data set requires a considerable amount of human involvement. In addition, the threshold-based evaluation systems that employ one or more thresholds while making correspondences between the detection and reference sets are also subjective since there is no unique way to choose the thresholds. This paper proposes an automatic and threshold-free evaluation system using the centre distances between the detected and reference entities. It neither requires any human involvement nor uses any thresholds. As a result, the proposed evaluation system can be exploited for a bias free evaluation of any roof reconstruction techniques, even on large data sets.

Automatic co-registration is a basic step in multi-sensor data fusion for remote sensing applications. The effectiveness of Mutual Information (MI) as a similarity measure for multi-sensor image registration has previously been reported for medical and remote sensing applications. In this paper, a new intensity-based approach built on local MI principles is presented. The approach decreases the complexity of higher dimension optimization by measuring local MI on well-distributed tie points. In addition, the reliability of registration is improved due to utilization of redundant observations of similarity. The performance of the proposed method for the registration of WorldView2 satellite imagery with LiDAR elevation and intensity data has been experimentally evaluated and the results obtained are presented.

Corner detectors have many applications in computer vision and image identification and retrieval. Contour-based corner detectors directly or indirectly estimate a significance measure (e.g., curvature) on the points of a planar curve, and select the curvature extrema points as corners. While an extensive number of contour-based corner detectors have been proposed over the last four decades, there is no comparative study of recently proposed detectors. This paper is an attempt to fill this gap. The general framework of contour-based corner detection is presented, and two major issues-curve smoothing and curvature estimation, which have major impacts on the corner detection performance, are discussed. A number of promising detectors are compared using both automatic and manual evaluation systems on two large datasets. It is observed that while the detectors using indirect curvature estimation techniques are more robust, the detectors using direct curvature estimation techniques are faster.

Precise georeferencing is one of the prerequisites for orthoimage generation from high-resolution satellite imagery. This requires the availability of a small number of GCPs that have to be visible in each scene. In this paper, it is shown how the number of GCPs required for the precise georeferencing of Advanced Land Observation Satellite (ALOS) imagery can be reduced by up to 90% using a generic pushbroom sensor model and strip adjustment while still achieving an accuracy of better than 1 pixel. A fully automatic work flow requiring an existing digital orthophoto and a digital elevation model (DEM) is also presented. Using an orthophoto mosaic generated from Landsat-7 panchromatic imagery for automatic GCP measurement, pixel-level accuracy can be achieved for images from the Advanced Visible and Near Infrared Radiometer type 2 (AVNIR-2) instrument. For images from the Panchromatic Remote-sensing Instrument for Stereo Mapping (PRISM), the accuracy of the automated procedure is only about 2 pixels due to the poor resolution of the orthoimage, compared with PRISM.

The Indian-ocean tsunami occurring on the morning of 26 December 2004 caused the most devastating disaster in the known history of Thailand. Remotely-sensed data, particularly high-resolution satellite imagery such as 1m Ikonos images, played a crucial role as an up-to-date post-event information source for a variety of user groups including rescue teams, relief and aid workers, and policy makers. However, ready-to-use low-level image products shipped by the provider can contain absolute positioning errors ranging from 5 - 30 m. Mismatches can therefore occur when georectified imagery is overlaid with vector data such as transportation networks or landmarks and this can complicate further analysis, both visually and analytically. More metrically precise image products can be generated through photogrammetric techniques, but systematic errors in the provided rational polynomial coefficients (RPCs), which describe the relationship between image and ground coordinates, can still cause computed absolute ground coordinates to be in error by several pixels. The bias-compensation technique described in this paper can be used to correct RFC values. Test results on stereo-pairs of Ikonos images have shown that the resulting horizontal error of bias-corrected imagery is reduced to just above the 1m level, making the image comparable in terms of metric accuracy to a 1:5,000 scale map. This 1m level accuracy can be achieved with only one ground control point, the coordinates of which are measured by the real-time kinematic (RTK) GPS technique. Furthermore, field control point measurements can be totally avoided if large-scale maps of the desired areas exist and well-defined points can be found to be present in both the map and image. The presented bias-compensation technique has clearly demonstrated its suitability as a rapid geopositioning correction tool for the generation of precise satellite image products for disaster management. © 2005 IEEE.

Sensor orientation based on a 3D affine transformation is appealing for high-resolution satellite imagery for three reasons: 1) Narrow-angle imaging sensors can be characterised by a skew parallel projection, 2) the affine model is linear, and 3) it is a straightforward computational model requiring a minimum of only four ground control points. Application of the affine model with Ikonos imagery has produced sub-pixel 3D geopositioning accuracy. In spite of this, there remain questions about the fidelity of the model where the area covered is large and where there is a significant terrain height variation. This paper describes the application of the affine model to multi-scene Ikonos stereo imagery covering large block areas. It is shown that area is not a factor influencing the impressive metric performance of the affine for pixel-level geopositioning.

This paper discusses a network for computer vision which comprises multiple nodes, each node indicating one object point and its corresponding multiple imaging rays. The motivation behind this multi-image network approach are threefold: 1) the use of redundancy to afford less reliance on scene intelligence; 2) the pursuit of quality through the quantity of information available in overdetermined systems; and 3) the treatment of networks of images rather than independent image pairs. The primary characteristics of a network vision scheme for 3D object reconstruction include application of epipolar curve constraints, use of multi-ray triangulation residuals in object space, adoption of least-squares network optimisation and application of global quality control measures. The matching speed for object point determination in the reported network vision implementation, comprising four images, reaches 120 points per second using a Pentium-200 processor. A 3D triangulation accuracy of close to 0.1 pixels is achieved.