U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
202-366-4000
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
| REPORT |
| This report is an archived publication and may contain dated technical, contact, and link information |
|
 |
| Publication Number: FHWA-HRT-17-048 Date: May 2018 |
Publication Number: FHWA-HRT-17-048 Date: May 2018 |
This chapter is organized into four sections: Study Design, Data Collection Process and Methods, Site Descriptions and Results, and Field Study Findings. For each section, individual sites and locations are addressed separately.
Based on the topics and related attributes identified in previous work products, the project team developed data collection activities for each of the 21 sites considered for field evaluations. The project team considered the collection of video, both aerial and from fixed-location cameras; the collection of photographs and observational notes; and site evaluations consisting of a rudimentary TCD audit and examination of the geometric configuration of the interchange decision points. The initial list was refined to six sites identified for field evaluations (table 70).
TH 62 = State Trunk Highway 62; UAV = unmanned aerial vehicle.
The following subsections discuss the site visit preparation, data collection methods, data reduction, and data analysis.
In preparation for the site visits, the project team examined all access points along each segment using satellite imagery. The team determined the ideal locations for fixed-position cameras and unmanned aerial vehicle (UAV) hovering in an effort to ensure that data collection methods would produce imagery without gaps or areas out of view. This preparation also enabled the team to write appropriate requests for proposals to vendors, ensuring that each vendor received the same information and that the camera positions were reasonable given the type and quantity of data desired.
Three types of data were collected for this project: photographs, videos from fixed-location cameras, and videos from UAVs.
Photographs
The field data collection for behavioral analysis was limited to representative sites with exceptional complexity. For many sites, the use of fixed-location cameras was cost-prohibitive due to the relative lack of structures for camera mounting and an unfavorable safety risk analysis for camera deployment activities. In the case of site 31, the use of UAVs was explicitly prohibited by the proximity of the Minneapolis-St. Paul International Airport and the instrument approach paths for runways 12L and 12R and departure paths for runways 31L and 31R. To collect photographs, the project team drove through the interchange being studied and captured images at guide signs and pavement markings.
Fixed-Location Camera Video
Fixed-location cameras were deployed at site 11 (supplementing the aerial video), site 27, and site 31 to collect field data. For sites 11 and 27, the project team used high-resolution cameras with a weather-resistant housing and supplemental battery. For site 31, a vendor was selected who provided equipment used for intersection traffic counts, including the installation, removal, data download, and delivery.
For all installations, project team and contractor members used personal protective equipment, including reflective vests, high-visibility headwear, safety glasses, and rated footwear. In addition, project staff conducted a risk analysis, including assessing vehicle parking and site access. No work plans involved the closure of lanes on the freeway or flagging activities related to vehicle diversions.
UAV Video
UAV video data were collected using high-resolution, 4K cameras mounted to professional-quality UAVs. Battery technology limited each UAV sortie to approximately 17 to 20 min, allowing for 15-min data collection intervals separated by approximately 5 min of transit and servicing time. The project team and contractor worked together in advance to create a collection plan, including arranging with property owners for permission to set up launch locations that permitted unobstructed views of the drones to comply with Federal Aviation Administration regulations. In addition, the entire team (operators, spotters, and support staff) met twice each day for a mission briefing, including analyzing risk, assigning and inspecting personal protective equipment, and establishing communications and command protocols.
The project team reduced all field video to facilitate data analysis. Data reduction activities varied per site and focused on distilling driving behavior (e.g., route choice, lane selection, origin/destination) from the video. In general, the data reduction activities involved viewing each video, selecting vehicles based on a predetermined sampling plan, and noting the lane position of each selected vehicle at screen lines along the route. Those screen lines were chosen based on the position of guide and regulatory signing, the location of pavement markings and pavement marking pattern changes, and other factors that could influence driver behavior relative to navigating to an exit. Where possible, qualitative observations concerning vehicle trajectory, speed, and sudden movements were also recorded and compiled to determine if some locations exhibited a higher frequency of these types of movements.
The project team primarily used two approaches to analyze the field data: (1) entering the data into a relational database and manipulating the data through queries and (2) cluster analysis.
The first approach was primarily used to drill down into the data and ask specific questions about how drivers behaved within the captured field video data. Example questions asked through this approach ranged from basic (e.g., what percentage of drivers used the option lane) to more complex (e.g., what percentage of vehicles from any non-exiting lane moved into an exiting lane and ultimately exited). Because the data coders reduced all field data in a uniform format (see figure 43), the use of relational databases allowed the project team to reuse queries at different sites after making minor tweaks to each query.
Source: FHWA.
A. Data examples from site 27 as provided by data coders.
Source: FHWA.
B. Site 27 data entered in the relational database.
Figure 43. Graphics. Site 27 data.
The second approach used for the data analysis was a cluster analysis. Cluster analysis is a data mining technique that groups data into clusters, in which the data items within a single cluster are similar, but data items between clusters are dissimilar. The project team used open-source data mining software that included an algorithm for k-means clustering.(43) All cluster analyses performed on the field data were completed using the simple k-means algorithm ranging from 2 to 10 clusters. The final number of clusters was selected by identifying the “elbow” of the curve when the numbers of clusters were plotted against the sum of error within each cluster; this is the point at which additional clusters do little to better explain the dataset. This analysis provided groups of common driving behaviors (i.e., what lanes drivers selected at different locations) witnessed in the field video data.
For this analysis, lane selection per location within a single study site was used on input. Output of the cluster analysis showed what lane best represented driver behavior at each location within the study site. An example of analysis output is shown in figure 44. It is important to note that this method assigns vehicles to the cluster that most closely represents their driving behavior. In other words, while the driver behavior will be similar between drivers of a single cluster, not every driver within a given cluster has the exact same driving behavior.
The following sections describe four sites: site 11 in Orlando, FL; site 27 in Atlanta, GA; site 31 in Minneapolis, MN; and site 43 in Olympia, WA.
The following subsections cover the description, observations, and results for site 11.
Description
A total of five UAVs were used to provide coverage of roughly 15,000 linear ft along I-4, between US 192 in the southwest and SR 536 in the northeast. The view of the area is depicted in figure 45. Of particular interest was the behavior of drivers entering from World Drive and US 192 in the eastbound direction and drivers entering from SR 535 and SR 536 in the westbound direction. The area is particularly complicated by the presence of braided ramps between World Drive and US 192, between US 192 and Osceola Parkway (County Road 522), and between Osceola Parkway and SR 536. This means that there is no I-4 access to US 192 from Osceola Drive in the westbound direction. This site was chosen particularly because it addresses the system design aspects of complexity.
The location of the UAVs varied according to the direction being filmed. The composite view of the UAVs provided the ability to track vehicles entering from World Drive southwest of US 192 and exiting as far north as SR 536 in the eastbound direction, while westbound vehicles entering from SR 535 could be tracked to the directional ramps on the US 192 interchange. An example image from the aerial video is shown in figure 46. The data collection took place on Thursday, February 11, and Friday, February 12, 2016.
The following attributes were present at the site:
Observations
Video was collected from both the eastbound and westbound approaches and will be discussed separately in the Results subsection. In both cases, the sample of vehicles selected for this analysis represents an equal distribution of vehicles across the three lanes of I-4. Approximately 70 min of video per UAV were reviewed for the eastbound video and approximately 50 min of video per UAV were reviewed for the westbound video. Vehicles were sampled at a rate of one sample per minute, with each sample representing one vehicle per lane. Vehicles were tracked across the entire study area using videos from each of the five UAVs, and any vehicle that could not be positively tracked across the entire study area was removed from the dataset. The total sample size resulting from the data reduction was 328 vehicles. More information on the distribution of exiting traffic is shown in table 71 and table 72.
Results
Results were obtained from both the eastbound and westbound approaches to site 11.
Eastbound
Figure 47, figure 48, and figure 49 illustrate common driver behavior identified at exits 64, 65, and 67, respectively, on eastbound I-4. Depiction 3 in figure 49 illustrates apparent ULCs where drivers move out of the option lane to exit the freeway in the exit-only lane, yet these drivers make a lane change to the left on the C/D roadway to exit left at the downstream split.
Source: FHWA.
Figure 47. Graphic. Common exiting driver behaviors at site 11, eastbound, exit 64 (n = 24).
Source: FHWA.
Figure 48. Graphic. Common exiting driver behaviors at site 11, eastbound, exit 65 (n = 6).
Source: FHWA.
Figure 49. Graphic. Common exiting driver behaviors at site 11, eastbound, exit 67 (n = 10).
Westbound
Figure 50, figure 51, and figure 52 illustrate common driver behavior identified at exits 64, 65, and 67, respectively, on westbound I-4.
Source: FHWA.
Figure 50. Graphic. Common exiting driver behaviors at site 11, westbound, exit 64 (n = 26).
Source: FHWA.
Figure 51. Graphic. Common exiting driver behaviors at site 11, westbound, exit 65 (n = 13).
Source: FHWA.
Figure 52. Graphic. Common exiting driver behaviors at site 11, westbound, exit 67 (n = 45).
The following subsections cover the description, observations, and results for site 27.
Description
The project team deployed five cameras along westbound I-20 near Atlanta, GA, on October 20, 2015. Approximately 4 h of video were collected, beginning approximately at 1:30 PM and ending at approximately 5:30 PM. Cameras were mounted on 20-ft telescoping masts and attached to roadside hardware. Each camera has an internal Global Positioning System (GPS) antenna that records GPS location, time, and date, which allowed the data reduction team to synchronize the collected video, enabling accurate correspondence of multiple camera angles.
Site 27 includes an approximately 1.5-mi stretch of I-20 eastbound leading up to exits 51A and 51B. In this location, I-20 eastbound consists of four lanes, with the rightmost lane an exit-only lane, and the adjacent lane an option lane at both the first (exit 51A) and second (exit 51B) exits. An aerial image of the study site is shown in figure 53, and an example image from the field video is shown in figure 54.
The following attributes were present at the site:
Observations
The sample of vehicles selected for this analysis represents an equal distribution of vehicles across the four lanes of I-20. Ninety min of video data were reviewed, beginning at about 4:00 PM. The video was sampled at a rate of two samples per minute, with each sample representing one vehicle per lane, making a total of four vehicles per sample. For instances where a vehicle was not present in a lane through the whole sample interval (30 s), no vehicle was recorded for that sample. The total sample size resulting from the data reduction was 719 vehicles, with 51 percent of the vehicles exiting I-20 over the course of the study area. More information on the distribution of exiting traffic is shown in table 73.
Results
Figure 55 and figure 56 illustrate common driver behavior identified at site 27. In addition, video coders identified several instances of erratic behavior by vehicles near exit 51B. Within the 90 min of field video reduced, two passenger vehicles were observed standing in the exit gore before returning to the freeway, and two commercial vehicles were observed crossing the exit gore. While one passenger vehicles only entered the exit gore briefly, the other remained in the exit gore for more than 2 min, rejecting many opportunities to return to the mainline. One commercial truck began to take exit 51B when the driver crossed over the exit gore to return to the mainline. The second commercial truck came to a complete stop on the mainline adjacent to the exit gore for several seconds before crossing the exit gore to take the exit.
The following subsections cover the description, observations, and results for site 31.
Description
Site 31 included three approaches to the interchange of I-35W with Minnesota State Trunk Highway 62 (TH 62, the “Crosstown Freeway”). These approaches included southbound I-35W, westbound TH 62, and northbound I-35W, in Richfield and Minneapolis. Of the three approaches, the southbound I-35W approach to TH 62 was the most complicated, with three total splits and a total of four downstream. The three interchanges are summarized in table 74.
| Location | Route | Direction | Interchange |
|---|---|---|---|
| 31-1 | I-35W | Southbound | TH 62 |
| 31-2 | TH 62 | Westbound | I-35W |
| 31-3 | I-35W | Northbound | TH 62 |
Site 31-1
Site 31-1 encompassed a distance of roughly 1½ mi along southbound I-35W. Project contractors deployed three cameras upstream of the split of the three-lane exiting roadway from I-35W southbound. The cameras were deployed according to the information in table 75.
At E 50th St, the northernmost location, cameras were mounted facing both north (against traffic) and south (with traffic, similar to cameras 2 and 3). The camera systems at site 31-1 were adequate to provide identification on vehicle color and type, but generally lacking the resolution necessary to provide positive vehicle tracking over long distances, partly due to the mounting height of the cameras, which was generally within 20 ft of the roadway surface and directly overhead. The data reduction efforts for this location and other locations within site 31 prompted the team to use digital, high-resolution cameras for other sites, as these cameras produced standard National Television System Committee imagery.
The first downstream split involves five lanes diverging into six (see figure 57), with the center lane serving as an option lane for the exit. The signing in advance of all these locations consisted of discrete arrow signing using a vertical divider above the option lane arrow, depicted in figure 58. The arrows at the departure location are angled to better emphasize the point of departure.
Source: FHWA.
Figure 58. Photo. I-35W southbound approaching split 1, the three-lane ramp to the Minnesota TH 62 exits.
The three-lane exit ramp diverges downstream (approximately 1,700 ft) into two two-lane ramps, as depicted in figure 59. This second divergence is referred to as split 2. Each of those two ramps further splits, with exclusive-lane exits to Portland Avenue (the left-hand ramp) and Lyndale Avenue (the right-hand ramp) and are built to accommodate two-lane entrances to TH 62 as the Crosstown Freeway is expanded in the future.
The project was primarily concerned with the behavior of drivers at split 1 and at split 2. While the behavior of drivers between the upstream guide sign and splits 3L and 3R is also of interest, the signing for those exits and pavement markings are thorough and conventional.
The following attributes were present at the site:
Site 31-2
An illustration of the camera installation locations is shown in figure 60. The first camera upstream (camera 31-2-1) was attached to the bridge railing on the Portland Ave. S overcrossing, and the second camera (camera 31-2-2) was attached to a ramp meter signal post in the vicinity of the ramp to I-35W northbound. Video from camera 31-2-1 depicts vehicles’ behavior at the section of TH 62 where an additional lane is added on the right, creating three lanes. Video from camera 31-2-2 depicts drivers’ destination selection (i.e., remain on TH 62 or exit to I-35W northbound).
The following attributes were present at the site:
Site 31-3
Two ground-mounted cameras were installed adjacent to an exit gore at exit 11 on I-35W near Richfield, MN (figure 61). Field video collected traffic behavior of vehicles as they exited I-35W and traveled on the exit ramp to TH 62.
©Esri.
Figure 61. Photo. Camera positions at site 31-3 (I-35W northbound approaching Minnesota TH 62).(47)
The single-lane exit ramp expands into two lanes after approximately 400 ft from exiting I-35W, and the ramp splits with the exit on the left bringing the driver to TH 62 westbound and the exit on the right bringing the driver to TH 62 eastbound.
The following attributes were present at the site:
Observations
The following observations were made at site 31-1, site 31-2, and site 31-3.
Site 31-1
For the video data reduction, the video was sampled twice a minute, and one vehicle from each lane was selected in each sample, resulting in the capture of driver behavior from 95 vehicles from the approximately 20 min of video. Video was sampled once per minute, collecting information on one vehicle per lane across the five lanes of I-35W, totaling 100 vehicles. Five vehicles were removed because they could not be positively identified at the second camera location. Basic information on the sample is shown in table 76.
| Number of Observations | Number/Percent Remaining on I-35W | Number/Percent Exiting I-35W |
|---|---|---|
| 95 | 62 (65%) | 33 (35%) |
Site 31-2
For the video data reduction, the video was sampled twice a minute, and one vehicle from each lane was selected in each sample, resulting in the capture of driver behavior from 474 vehicles from the approximately 2 h of video. Approximately 1 h of video was recorded during the AM peak period, and approximately 1 h was recorded during the PM off-peak period. Basic information about the traffic during the two times is included in table 77.
This site focused on understanding vehicle behavior at a location upstream of an exit where the freeway expands from two to three lanes. The project team recorded vehicle position immediately prior to the addition of the third lane and after the third lane was fully established. Finally, the vehicles’ destinations (e.g., remain on the roadway or exit the freeway) were recorded.
The hypothesis was that drivers intending to exit to I-35W north would avoid use of the right-hand lane until the overhead signing was visible. Some drivers may use the lane immediately in heavy traffic or because their intended destination is the next right-hand exit on I-35W north.
Site 31-3
Driver behavior of 2,144 vehicles was captured from approximately 3 h of field video. One h of video was reviewed during the AM peak, 1 h was reviewed during the PM off-peak, and 1 h was reviewed during the PM peak. Basic information about the driver route selection during the three time periods is included in table 78. PM off-peak and PM peak times were selected to maximize the percent of traffic exiting to TH 62 East.
The study area was divided into three regions to observe how vehicles behaved when approaching the split. The first region captured the vehicle’s orientation as it entered the study area; the second and third regions approximately divide the remaining exit ramp in two, with the first region (region 2) being before the driver encounters the lane designation pavement markings. Finally, the project team recorded the destination selected by each driver.
The hypothesis at this location was that vehicles exiting to TH 62 eastbound would move into the formed right-hand lane further down than the beginning of the lane addition taper, owing to the lack of upstream guide signing with route marking. Further, it was hypothesized that vehicles would drift toward the right, following the right white edgeline, on account of the lack of dotted extension lines providing positive guidance and indicating to drivers the lane addition taper and the path for those wishing to remain in the left-hand lane. This anticipated drifting is a characteristic behavior on wider freeway exit ramps and, without information concerning the arrangement of the lanes, can occur with symmetrical and asymmetrical widenings.
Results
Figure 62 illustrates common driver behavior identified at site 31-1.
Findings from site 31-2 show that fewer drivers exiting the freeway entered the newly formed exit lane near the beginning of the lane (i.e., once it first became available) than those that entered it later. This finding could indicate that exiting drivers are changing lanes further downstream (i.e., closer to the exit) because of uncertainty in identifying the proper lane to navigate the exit.
Findings from site 31-3 found that traffic exiting to the left (westbound) displayed more uniformity than the traffic exiting to the right (eastbound). Eastbound traffic exhibited three typical behaviors. The most common behavior for the eastbound traffic was to enter the study area in the center of the lane, maintain the position in the center of the lane, and then move to the right side of the lane near the exit. Fewer eastbound vehicles entered the study area on the right side of the lane and maintained that position before exiting to the right.
In addition to the findings from the cluster analysis, the video reviewers provided qualitative descriptions of the field video from site 31-3. In the 3 h of video, several instances of erratic or uncertain behavior were identified. During the PM off-peak time period, five vehicles were observed exiting the freeway by crossing the exit gore. During the same time period, one vehicle was identified crossing the exit gore at the downstream split. During the PM peak time period, one vehicle was observed crossing the exit gore, and another vehicle was observed missing the exit, pulling over to the shoulder, and backing up to take the exit.
The following subsections cover the description, observations, and results for site 43.
Description
Similar to site 11, site 43 featured a length of urban freeway with longer spacing between exits, across a distance of 7 mi. Along I-5, three interchange locations were selected, summarized in table 79. Each exit features an option lane from the mainline roadway.
Two UAVs were launched for each sortie on Tuesday, November 5, 2015, for each of the three interchange locations in site 43. Because of favorable weather and lighting conditions, field video captured driver behavior on roughly more than 4,000 linear ft of I-5 in advance of each of the three selected interchanges. The field video captured the behavior of drivers as they approached each individual interchange.
Site 43-1
Site 43-1 includes slightly less than 1 mi of I-5 southbound focusing on exit 105B and exit 105A. In this location, I-5 southbound consists of four lanes, with the rightmost lane an exit-only lane, and the adjacent lane an option lane. An aerial image of the study site is shown in figure 63, and an example image from the aerial video is shown in figure 64.
The following attributes were present at the site:
Site 43-2
Site 43-2 includes slightly less than 1 mi of I-5 southbound focusing on exit 104. In this location, I-5 southbound consists of four lanes, with the rightmost lane an exit-only lane, and the adjacent lane an option lane. An aerial image of the study site is shown in figure 65, and an example image from the aerial video is shown in figure 66.
The following attributes were present at the site:
Site 43-3
Site 43-3 includes slightly less than 1 mi of I-5 northbound focusing on exit 111. At this location, I-5 northbound consists of three lanes. Approximately halfway through the study area, a fourth lane is added with the rightmost lane an exit-only lane, and the adjacent lane an option lane. An aerial image of the study site is shown in figure 67, and an example image from the aerial video is shown in figure 68.
The following attributes were present at the site:
Observations
Driver behavior was captured at three locations at site 43. Basic information on driver route selection at each location is presented in table 80.
Results
Figure 69, figure 70, and figure 71 illustrate common driver behavior identified at site 43.
The field study explored driver behavior at complex interchanges in several areas across the country. Field video data were collected at 13 interchanges spanning 4 States through the use of both fixed-position cameras and UAVs. For each interchange studied, field video was reduced to track vehicle paths throughout the study site. While data were captured on both through and exiting traffic, the focus of this review was on the exiting vehicles. Because exiting vehicles served as the focus of the analysis, in many cases, sample sizes were relatively small and data coding was time consuming, particularly when combining several camera views.
One common finding across sites is that exiting traffic was found to most commonly use the exit-only lane rather than the option lane. The cluster analysis found only one site where the most common group of driver behavior did not include moving to the exit-only lane before, or near, the first guide sign with a yellow “EXIT ONLY” panel. The analysis also identified examples where drivers would use the exit-only lane and, ultimately, make a lane change on the C/D roadway. These instances include signing that does not indicate the presence of an option (e.g., site 11 eastbound) lane or groups several destinations on a single sign (e.g., site 43-1). Conversely, the cluster analysis did not find any instances of common behavior where drivers exited using the option lane, and then changed lanes on the C/D roadway.
The site with the highest option lane use was found to be Washington State site 43-2. At this site, nearly half of the drivers were observed using the option lane. This site also included a guide sign configuration not seen in any of the other sites. On the advance guide sign, the option lane is signed with the standard downward white arrow on a green background; however, the guide sign also includes the letters “OK” adjacent to the downward arrow.
Few common behaviors identified through the field video show last-minute lane changes. Behaviors with last-minute lane changes were found in the Atlanta, GA, site (both exits 51A and 51B), as well as Washington State site 43-3. In all cases, the percentage of vehicles completing these lane changes was small. Further, for the case of exit 51B in Atlanta, these lane changes may be influenced by interchange geometry as well as (or in lieu of) signing because exits 51A and 51B are separated by slightly more than ¼ mi.
There was one instance of a common behavior found that could be interpreted as a misunderstanding of the guide signs. In the Atlanta, GA, site, about 12 percent of the drivers exiting at the second exit (51B) were classified in a cluster of driving behaviors that showed the drivers traveling in the exit-only lane for the first exit (51A) before changing out of the lane near the exit ramp for exit 51A to proceed to exit 51B. Drivers who followed this behavior passed two sign gantries where the sign designated the lane as an exit-only lane for exit 51A, with the first gantry also having a separate advance guide sign for exit 51B.
