Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
1Overview Linking the second Strategic Highway Research Program (SHRP 2) Naturalistic Driving Study (NDS) data to the SHRP 2 Roadway Information Database (RID) will support research- ers whose work involves relating information from the driving data to information in the roadway data. The SHRP 2 project collected more than 1 million hours of driving data: it mea- sured speed, acceleration, latitude and longitude, distance and speed relative to other vehicles, as well as recorded video of the vehicleâs driver and its surroundings. Additionally, the project collected roadway data and made detailed descriptions of a subset of those roads driven, such as the number of lanes, pres- ence of rumble strips, barrier types, and shoulder width. Almost all driver- and vehicle-related safety research questions benefit from considering the surrounding roadway environ- ment, and most roadway safety research questions benefit from the study of drivers and vehicles on comparable sections of roadway. Many believe the highest return on safety research will be achieved through the study of the interactions among the driver, roadway, and vehicle. This project is intended to provide researchers with a means for connecting or associating the data collected from the driver and vehicle subsystems and the roadway subsystem. As described in detail below, that âmeansâ is a massive linking table (approximately 305 million rows with 2.6 million unique links) containing a listing of map-based roadway links traveled by each NDS vehicle on each trip it made. The RID contains the same map-based link identification numbers for each roadway segment for which inventory data have been collected. The number of links for which data are available varies, depending on the type of data. For example, approximately 120,000 links have measures col- lected by roadway data collection vans. To deliver this association, the vehicle position, reported by a Global Positioning System (GPS) once per second, had to be associated reliably with the roadway on which it was travel- ing. Throughout this task report, the processes used to make this association are referred to as the Matching Process, or the matching of GPS points to the roads on which a vehicle has traveled. Figure 1.1 illustrates accurate GPS data collection from a vehicle overlaid on a digital map. The reported positions agree well with the location of the roadways shown on the underlying aerial imagery and the digital map. Through visual inspection, one can interpret the GPS points relative to the digital map data and identify the traveled roadway. Figure 1.2 provides an example in which the reported GPS location of the vehicle deviates quite far from the roadway. The accuracy of the position reported by GPS varies depend- ing on the orientation of the visible satellites to the vehicle. This project applied a data processing algorithm, known as the Matching Algorithm, that interpreted GPS points and digital maps such as those shown in Figure 1.1 and Figure 1.2, identi- fied the actual road traveled, and then recorded the traveled roadway in a database. Previous Methods for Associating GPS Points with Roadways Using GIS-Based Tools The digital maps defining roads are made up of many line seg- ments, referred to as links. Links are connected to each other at nodes. Nodes are present at intersections but are also found at other locations along a section of roadway. Later in the report, Figure 1.3 and Figure 2.2 show further details of these digital maps. Buffers Previous methods for associating GPS points with roads have used geospatial functions such as buffers or spatial joins (Wu and McLaughlin 2012) within a geographic information sys- tem (GIS). Buffers work by defining a buffer area around the C h A P T e R 1 Background
2no node where Link DE crosses Link WX, since Link DE has an overpass over Link WX. This indicates on the digital map that a vehicle cannot travel from Link DE directly to Link WX. GPS latitude and longitude pairs captured by the vehicle are shown as green dots traveling from Point A to Point F. Also, near Node C, a GPS point is recorded on Link CU. Similarly, a GPS point lands on Link WX and Link EZ (see Figure 1.3). If a buffer is drawn around each of the links and captures the GPS points within the buffer, the solution would indicate the vehicle has traveled on all the correct links but would also incorrectly indicate that the vehicle traveled on Links CU, WX, and EZ. This solution is portrayed in yellow in the right panel of Figure 1.3. Spatial Joins An alternative to a buffer approach is to use a spatial join. A spatial join is a geospatial database method that associ- ates objects in space based on their proximity. A spatial join processes roadways as two- or three-dimensional objects through space. The GPS points are also objects in that space. link, such as 50 ft on each side of the link, and processing GPS points relative to the buffer. GPS points landing inside the buffer are associated with the link around which the buffer was drawn. Points outside the buffer are either not assigned to a roadway or are assigned to some other roadway if they fall within the buffer around that roadway. There are several shortcomings of this approach for deter- mining what road a vehicle was on. First, it is difficult to know how large to make the buffer. As we can see in Figure 1.2, the reported position can wander significantly from a road. If the buffer is too small, GPS points will be ignored. If the buffer is too large, GPS points might be assigned to the wrong road. Second, where roads intersect, the buffers will overlap. For example, the left side of Figure 1.2 shows GPS points from a vehicle traveling east to west that land on a road intersecting from the south. A buffer approach would not be able to deter- mine which road to assign these points to. Figure 1.3 shows an example of the results of this buffer type of method. The left panel of the figure portrays a digital map with lines showing the path of the roads; nodes are marked at intersections with black circles and letters. Note that there is Figure 1.1. Illustration of GPS points that follow a roadway. Figure 1.2. Illustration of GPS points that deviate from a roadway and land on an intersecting roadway.
3 The join computes the distance between the roadways and the GPS points and assigns GPS points to the nearest roadway. Unfortunately, this approach also has limitations. Though this method does not require setting a buffer size ahead of time, it still suffers a similar limitation where roads intersect. If a GPS point lands closer to an intersecting road than to the actual road of travel, it will be assigned to the intersecting road, and the solution will result in errors similar to those shown in the right panel of Figure 1.3. A second limitation of this method is that it is computationally intensive. The roads in the digital map have shape over their length through the space. Comput- ing what part of millions of miles of digital roadway shapes are closest to approximately 3.7 billion GPS points in the SHRP 2 work is a demanding task even for high-performance comput- ing environments. In the end, the solution will still falsely identify roads that are near a GPS point but not actually the road on which the vehicle was traveling. Improving the Methods Further enhancements to these fundamental GIS methods are able to improve the accuracy of the output. Li et al. (2014) used a method of joining sequences of GPS points into a âtrip line,â and then associating these extended lines to nearby links. In the same work, road names of the roads near trip lines were used within the logic to identify road names that were near at least two of three points defining the line. Those roads that appeared more than twice associated with a trip line were considered the links on which the vehicle had trav- eled, and the GPS points defining the line were associated with that set of links. These methods increased accuracy beyond the fundamental GIS techniques and are candidates for use with small data sets, but they still have problems with false classifications and cannot be used to handle data sets on the scale of the SHRP 2 NDS. Figure 1.3. Illustration of a route solution based on GPS points relative to roadway spatial join or GPS buffer solutions.
