Executive Summary he Linear Referencing Profile (LRP) is part of the Location Referencing Message Specification (LRMS), a partial solution to interoperability problems in location expression and exchange (LX) in Intelligent Transportation Systems (ITS). This report is an evaluation of the LRP, carried out by the This document assumes that the reader is familiar with the background to interoperability problems in ITS, and the LRMS effort. The following points clarify the scope of this research: • Linear Referencing is most popular in GIST (Geographic Information Systems for Transportation), however, we treat the LRP as a generic ITS location messaging profile, not constraining our test approach to GIST scenarios. • We do not assume that the original expression of position is accurate; we examine the error in the process by which a linear reference is derived from 2-dimensional coordinates, and the error in measuring a linear offset using a Distance Measuring Instrument (DMI). • The LRP relies on indices for road sections and intersection nodes, that are common between users. If users communicate with respect to the same database, there is no interoperability problem; if the databases are different, then the task of assigning common identifiers is expensive, and some errors are inevitably introduced in this step. However our testing assumes away this problem, and we do not attempt to model or to estimate such error. We expend considerable manual effort to develop an accurate table of correspondences for a small sample of data, and we caution that in practical implementation, creating this table with requisite accuracy, and overcoming the inherent semantic conflicts (single vs dual line freeways, traffic circles, etc) will be a potential problem. Testing is built around three sets of experiments. The first involves field surveys using differential Global Positional Systems (GPS) and a Distance Measuring Instrument (DMI). In the preliminaries, we observe various characteristics of DMI readings under normal traffic conditions, and in remote test areas, to estimate the accuracy of the instrument and to quantify its limitations. Then a small sample of roads is selected; we drive those roads, and compare our readings to length calculations and coordinates from digital maps. One of those maps is an engineering scale product; measurements off the map are almost identical to our observations, both in terms of road lengths (±12m) and coordinates (±2m). Other maps disagree by 60–130m in length, and some coordinates are substantially in error (±200m). These numbers …
22 Figures and Tables
Table 1. The Linear Referencing Profile
Figure 1. A GPS point intended to snap to a1 may snap to a2 instead
Table 2. Principal classes of error in location exchange using LRP
Figure 2. Sample of 15 roads (they are distributed over a wide geographic area in reality; positions are distorted for this illustration).
Figure 3. Linear Transformation. Point I is surveyed by GPS, resulting in J, which snaps to K in the reference database B. Two cases of K are shown; one is clearly not on the intended road.
Figure 4. LRP Transfer Errors. Point P should ideally transfer to Pc; instead it transfers to Q using absolute offset, and R using normalized offset
Table 4 . Effect of lange changes
Table 5. Summary of DMI errors (worst case)
Figure 5. Worst case scenario of GPS error reversing itself at every observation
Table 6. DMI vs lengths measured from coordinates
Figure 6. Overlay of maps A, C and E, and GPS data (bold). C is almost completely obscured due to its agreement with GPS. E is grossly inaccurate, but shape points are dense and curves are smooth. A is relatively true but generalized.
Figure 7. Absolute Error (longest minus shortest) as a function of Road Length
Table 7. Success rates for coordinate snapping to intended section of intended street. All percentages are calculated with respect to the full set of test points, therefore “Correct Road” has the highest figures.
Figure 8. Relative Deviation (std devn/mean) as a function of Sinuosity
Table 8. Offset discrepancies, Test Set III
Figure 9. Shape points in database C are weeded out to produce C', which has the same number of shape points as A. The length of C’ is 7412m, still higher than A’s 7075m.
Table 9. Offset discrepancies, broken down by GPS error class.
Figure 10. Overlay of maps C (bold) and E, showing different positions (c and e) where ramp meets freeway. The impact on freeway segment length is more than 200m.
Table 10. Offset and 2-D errors, all roads
Table 11. Offset and 2-D errors, freeways only
Figure 11. Absolute offset as a function of magnitude of offset
Table 12. Offset and 2-D errors by sinuosity
Download Full PDF Version (Non-Commercial Use)