ASPRS

PE&RS June 2009

VOLUME 75, NUMBER 6
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
JOURNAL OF THE AMERICAN SOCIETY FOR PHOTOGRAMMETRY AND REMOTE SENSING

Peer-Reviewed Articles

657 Accuracy Evaluation and Sensitivity Analysis of Estimating 3D Road Centerline Length using Lidar and NED
Hubo Cai and William Rasdorf

Abstract  Download Full Article
Highway networks are represented by linear spatial objects (road segments). Having accurate length information of road centerlines is critical in transportation. This paper presents a geographic information system (GIS)-based approach that overlays planimetric road centerlines and elevation data to model road centerlines in a 3D space and estimate their lengths. Elevation sources included light detection and ranging (lidar) and the National Elevation Dataset (NED). The estimated distances were compared to distance measurement instrument (DMI)-measured distances to evaluate the accuracy. The effects of elevation datasets with varying vertical accuracies were assessed. The relationship between road geometric properties and the accuracy of distance estimates was examined. We found that (a) the proposed 3D approach is efficient in estimating 3D road centerline distances, (b) using lidar point data improves the accuracy by 28 percent over the use of NED, and (c) certain road geometric properties have direct relationship with the accuracy of distance estimates.

667 A Region-based Level Set Segmentation for Automatic Detection of Man-made Objects from Aerial and Satellite Images
Konstantinos Karantzalos and Demertre Argialas

Abstract  Download Full Article
A region-based level set segmentation was developed for the automatic detection of man-made objects from aerial and satellite images. The essence of the approach is to optimize the position and the geometric form of an evolving curve, by measuring information within the regions that compose a particular image partition based on their statistical description. The present region-based variational model is fully automated without the need to manually specify the position of the initial contour. Furthermore, it converges after a small number of iterations, allowing real-time applications. The developed algorithm was tested for the detection of roads, buildings and other man-made objects in a number of aerial and satellite images. The effectiveness of the algorithm is demonstrated by the experimental results and the performed qualitative and quantitative evaluation.

Color Figures 1 to 12 (Adobe PDF Format 743Kb)

679 An Adaptive Thresholding Multiple Classifiers System for Remote Sensing Image Classification
Yu-Chang Tzeng, Kou-Tai Fan, and Kun-Shan Chen

Abstract  Download Full Article
A multiple classifiers system which adopts an effective weighting policy to combine the output of several classifiers, generally leads to a better performance in image classification. The two most commonly used weighting policies are Bagging and Boosting algorithms. However, their performance is limited by high levels of ambiguity among classes. To overcome this difficulty, an adaptive thresholding criterion was proposed. By applying it to SAR and optical images for terrain cover classification, comparisons between the multiple classifiers systems using the Bagging and/or Boosting algorithms with and without the adaptive thresholding criterion were made. Experimental results showed that the classification substantially improved when the adaptive thresholding criterion was used, especially when the level of ambiguity of targets was high.

Color Figures 2 to 4 (Adobe PDF Format 100Kb)

689 Meta-Prediction of Bromus tectorum Invasion in Central Utah, United States
Nicholas Etienne Clinton, Peng Gong, Zhenyu Jin, Bing Xu, and Zhiliang Zhu

Abstract  Download Full Article
Cheatgrass (Bromus tectorum) is an invasive, exotic grass infesting the Western US. Multi-temporal Landsat TM imagery and ancillary topographic data were used for mapping this invasion over portions of Utah. Tobit, logit, probit, and Projection Adjustment by Contribution Estimation (PACE) regression, neural networks, and additive regression of regression trees were tested individually, and in an ensemble, Tobit regression had the best performance as an individual predictor. Tobit was most frequently the best predictor of zero cheatgrass coverage. A meta-predictor (classifier) to choose the best predictive model was implemented on a pixel-by-pixel basis. A J48 classification tree as a metapredictor resulted in an increase in accuracy over the best performer in the ensemble. This study illustrated the potential for meta-prediction as a general technique for increasing accuracy from a collection of base predictors.

703 Occlusion-based Methodology for the Classification of Lidar Data
Ayman F. Habib, Yu-Chuan Chang, and Dong Cheon Lee

Abstract  Download Full Article
Lidar systems have been widely adopted for the acquisition of dense and accurate topographic data over extended areas. The level of detail and the quality of the collected point cloud motivated the research community to investigate the possibility of automatic object extraction from such data. Prior knowledge of the terrain surface will improve the performance of object detection and extraction procedures. In this paper, a new strategy for automatic terrain extraction from lidar data is presented. The proposed strategy is based on the fact that sudden elevation changes, which usually correspond to non-ground objects, will cause relief displacements in perspective views. The introduced relief displacements will occlude neighboring ground points. To start the process, we generate a digital surface model (DSM) from the irregular lidar points using an interpolation procedure. The presence of sudden-elevation changes and the resulting occlusions can be discerned by sequentially checking the off-nadir angles to the lines of sight connecting the DSM cells and a pre-defined set of synthesized projection centers. Detected occlusions are then used to identify the occluding points, which are hypothesized to be non-ground points. Surface roughness and discontinuities together with inherent noise in the point cloud will lead to some false hypotheses. Therefore, we use a statistical filter to remove these false hypotheses. The performance of the algorithm has been evaluated and verified using both simulated and real lidar datasets with varying levels of complexity.

713 Isomorphism in Digital Elevation Models and Its Implication to Interpolation Functions
Peng Hu, Xiaohang Liu, and Hai Hu

Abstract  Download Full Article
Terrain is an ordered surface where locations relate to each other through their elevations. Preservation of this topographic orderliness by an interpolation method so that “if point A is higher than point B in the terrain, the interpolated elevation of A remains higher” is important to assure the practical value of the resultant DEM. Based on the concepts in group theory, this paper points out that a DEM must be isomorphic to terrain surface in order to preserve the topographic orderliness. Such a DEM, if generated through interpolation, can only be obtained if the interpolation function is an isomorphism. Two necessary conditions for isomorphism are identified: a one-to-one relationship between the topographic surface and the surface corresponding to the interpolation function, and the feasibility to configure the topographic surface into monotonic patches during interpolation. The isomorphism of three widely used interpolation methods is examined. It is found that linear interpolation in 1D and TIN are both isomorphisms, meaning that a DEM interpolated by either method can preserve the topographic orderliness. In contrast, bilinear interpolation is not an isomorphism. Considering the practical challenges in assuring the necessary conditions of isomorphism in each method, linear interpolation in 1D is recommended as the optimal method to interpolate a DEM.

723 Radiometric Normalization of SPOT-5 Scenes: 6S Atmospheric Model versus Pseudo-invariant Features
Aurélie Davranche, Gaëtan Lefebvre, and Brigitte Poulin

Abstract  Download Full Article
We compared the efficiency and robustness of two radiometric correction techniques applied to six SPOT-5 scenes used for assessing environmental changes of Camargue wetlands: the 6S atmospheric model and 86 pseudo-invariant features (PIFs) found in deep water, pine trees, roofs and sand. The few PIFs were selected subjectively following the low number of potentially invariant sectors available on the scenes. Both approaches provided a similar radiometric variation (6S = 4.3 percent; PIFs = 4.0 percent). The latter increased from water to pine trees, to roofs and sand, with five reference points per feature being identified as cost effective. The withdrawing of variant features among the PIFs across dates or points caused a significant decrease in radiometric variation, especially with 6S (6S = 2.8 percent, PIFs = 3.4 percent). As many as 31 point per type of PIFs would be necessary to provide a radiometric variation that is not significantly different from that obtained with 6S, whereas nearly 300 and 4,000 points per feature would be required to provide similar or better results than the 6S code, respectively. Use of a few PIFs remains a valid approach, as long as the invariant sectors cover a wide range of brightness and are represented by objects of which the radiometric variation has preliminarily been tested.

Top Home