Meetings — CLEM2001

Centerlines are required in a broad range of applications, from market research (where 1:100,000 scale and 50m error will do) to school busing and election planning (where it's important to know on which side of a street an address is), emergency services (20m) and precision snow plowing (10cm).  There are a variety of centerline mapping methods: field survey at the centimetre level, collaging CAD plans, GPS (5-20m), remote sensing and digital conversion of maps at different scales.  There is no universally "right" method, but there are application niches associated with each approach.  Some are faster, some cheaper, some more accurate; some apply to urban streets but not rural; some are compromised by urban canyons or tree canopies.  CLEM2001 is an opportunity to bring together some of the leading researchers in the field, to see if we can learn from each other and expand the domain of our solutions.

While everybody would like a street database "as accurate as possible", few people have spent much time systematically measuring positional accuracy of datasets such as TIGER or commercial street files.  This situation is exacerbated by huge variations in positional accuracy of individual components in evolving datasets such as TIGER, and because we happen to be on a cusp between using scale-based descriptions of accuracy such as the National Map Accuracy Standard (NMAS) and the recently adopted (1998) National Standard for Spatial Data Accuracy (NSSDA).  This paper describes Geographic Data Technology's experience in measuring accuracy of TIGER and GDT's proprietary "Dynamap" dataset.  It presents results of exhaustive and continuing positional accuracy tests of street centerline files before and after positional accuracy enhancement using a variety of techniques and sources.

The U. S. Census Bureau, in cooperation with the U. S. Geological Survey, developed the nations first street centerline file (TIGER) for use in the 1990 decennial census.  The Census Bureau is embarking on a program to modernize the MAF/TIGER system to support the use of GPS-enabled hand-held devices by its field staff.  The Census Bureau believes it requires a street centerline file and a coordinate for every housing unit at 3-meter or better accuracy to support this objective.  The Census Bureau has two projects underway utilizing imagery to realign TIGER/Line files and obtain structure coordinates.

Research on automated and semi-automated extraction techniques of linear features from remote sensing imagery has been active for decades.  Features of interest include transportation networks, power transmission lines, etc.  This paper presents a comprehensive survey of extraction techniques of such features from aerial and satellite imagery.  The techniques are evaluated with respect to methodology, strengths, drawbacks, and implementation approach.  Source data for the surveyed techniques include panchromatic and multispectral imagery.  The viability of hyperspectral data is extrapolated for same purpose of utilization.  The paper later presents a discussion of automated extraction techniques specifically used for updating road spatial databases.

Road network extraction from high-resolution images has significant applicability in transportation. It provides a means for creating, maintaining, and updating transportation network databases for many different purposes, such as infrastructure management, traffic safety analysis, and traveler information and guidance. This paper presents an automated procedure for road network extraction using optimization and supervised classification techniques. In order to establish a basic understanding of existing methods and approaches, the paper first summarizes important research efforts on the subject of road extraction. Then it provides a brief discussion on road image characteristics, focusing on road image intensity, image texture and geometric characteristics, which serves as a modeling aid for the computer-based extraction. After that, a road extraction method that uses dynamic programming and supervised classification is described, and some experiment results are illustrated.  Finally, strengths and shortcomings of the proposed method and some future research directions are discussed.

The centerline extraction problem may be cast in an information theoretic context. Two general classes of information are available from any image: spectral and spatial.  We demonstrate a novel filter technique that utilizes both classes of information. Spatial analysis on-the-fly is carried out by building a q-tree data structure on a pixel-by-pixel basis. This structure represents the local topology of the road network. Connectivity of pixels is determined by a matched filter distance metric using real spectra obtained by Dar Roberts and Meg Gardner of UCSB. In this way the technique combines spatial and spectral information to extract centerlines.

Under funding from US DOT, remote sensing researchers and software engineers at Veridian System have refined change detection and feature extraction software for preparation and maintenance of transportation databases.  The resultant tools have been incorporated into Classification and Feature Extraction - Transportation (CAFE-T), an adaptation of Veridian's general-purpose ERDAS Imagine-based CAFE toolbox.  The Change Detection tools are typically used to screen large areas using moderate-resolution imagery such as Landsat (15m to 30m; 7 spectral bands).  The feature extraction tools, on the other hand, target areas exhibiting high-densities of changes with higher-resolution multispectral imagery such as IKONOS (1m to 4m; 4 spectral bands) for improved characterization of the changes as well as extraction of new or modified road network segments.  Although the CAFE tools provide improved performance over traditional manual methods in the identification of land changes and associated road segments, the extraction and vectorization of these features does not currently offer much additional benefit over manual extraction.  Through collaboration with other transportation remote-sensing researchers, it is hoped that the efficiencies of the vectorization and clean-up routines can be improved to levels similar to these associated with the identification portion of the process.  If this can be accomplished, the resulting tools will significantly benefit large-area transportation database maintenance efforts across the country.

The capacity of remote sensing to produce quality data is running ahead of both the understanding and ability of transport route planners to use it.  The conventional road/rail route planning process is predominantly intuitive, and the weakest link is the planner's capacity to consider all data and the numerous constraints.  The Quantm system provides a method to unlock significant additional value from the data through its comprehensive analytical capabilities.  The result is alignments that deliver better engineering and environmental outcomes, and significantly reduce construction costs.  The magnitude and consistency of cost reductions are redefining the way that Australian road agencies consider data, with the quality and scope of remote sensing data becoming more attractive.

Kinematic GPS surveys offer major benefits to developing countries when it comes to characterising their network for management purposes. For the first time it is possible to quickly and cost effectively establish a road centreline. Over the last 18 months the World Bank has funded kinematic GPS centreline surveys in the Philippines, Lao PDR and Samoa. This paper will present the results of the surveys. It will address the technical issues -- such as differential corrections and projections -- as well as the operational issues: how does one execute a survey when the size and extent of the road network is in many respects an unknown. It will also discuss the larger issue of data accuracy: what is required for effective road management in developing countries.

The Ohio State University currently is developing a GPS/INS/CCD integrated system for precise  (centimeter level) monitoring of highway center and edge lines. This research is sponsored by the Ohio Department of Transportation. The prototype positioning component of the system is based on a tightly integrated GPS/INS system, and the imaging component comprises a single down-looking, high-resolution digital camera. The high image rate provides sufficient overlap of the subsequent images at highway speed, therefore stereo data processing is expected to be performed in real-time with the support of on-the-fly navigation solution.

This presentation will outline the use of vehicle probe data to achieve 100% accurate sub-meter map databases for all roads in US, Canada, and Europe that will be in use by the end of this decade.  This data can be made available by the vehicle manufacturers to help the needs of the infrastructure managers.  The probe data will be provided by vehicles from their advanced sensors for traction control, stability, night vision, adaptive cruise control, collision avoidance, road/lane departure prevention, etc.  This data will be collected by the vehicle manufacturers, assimilated, and broadcast back to vehicles as part of their real-time picture of the road around them for use by active safety products.

Multi-centerline representation of the divided highway provides more details and facilitates many transportation applications. However, the complexity of multi-centerline representation makes it unsuitable for network-related analyses, such as routing and some types of traffic analysis.  From a transportation engineering perspective, the paper discusses issues related to multi-centerline representation. It addresses issues encountered in practice through techniques developed using a combination of commercial software, in-house programs, and database queries. Issues include derivation of a single centerline from double centerlines, relating single and multiple centerlines, and standardization of topology directions. Finally, the paper concludes with the observation that it is ideal to include both representation schemes into a suite of interoperable GIS-T base maps and have them work together to satisfy specific needs of transportation applications.

Although recognized as a viable methodology, conventional aerial photography coupled with the photogrammetric process has generally been ignored for widespread, frequent roadway data collection because of cost considerations.  To enhance interest in traditional aerial photography and photogrammetry, we propose the following application to extract accurate roadway centerline and feature data.  Although initial costs are significant, it will be demonstrated that leveraging costs combined with a geospatial marriage of x,y,z coordinates with the transportation linear referencing system will make the process beneficial to multiple users.  A proprietary tool enhances the centerline and feature extraction process, and allows integration with van videologging data and stereo imagery from a desktop personal computer.