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.
8C h a p t e r 2 The Site Observer design and development ultimately are tied to safety research questions. Unlike vehicle-based naturalistic driving studies, the Site Observer is particularly capable of cap- turing objective data from a variety of vehicle-to-vehicle con- flicts, and because the most technically challenging scenarios relate to intersection safety, this application is the main focus of the study. Some consideration is also given to road depar- ture safety because this is also a major theme of the SHRP 2 Safety program. However, the topic of road departure safety is a more natural one for vehicle-based data collection, so the need for the Site Observer to address this is less clear. In the future, other safety areas, such as lane change and merg- ing conflicts, may prove amenable to analysis using auto- mated video tracking, but these are not considered in the current study. Intersection Safety The ability to predict or model the occurrence of crashes at intersections has been a challenge to transportation engineers since the early days of motorized transportation, and there have been many empirical and theoretical efforts to model intersection crash occurrence. Most studies have used traffic volumes as exposure measures and either implicitly or explic- itly incorporated risk into the models. The configuration of the intersection, the number of different types of turns that can be made, the type of traffic control, and driver compli- ance with the traffic control determine the encounters in which various types of crashes can occur. Early studies of intersection crashes found that the intersec- tion crash rate per volume of traffic was sensitive to changes in the proportion of traffic flow from the various legs of the inter- section. An early and widely known study by Tanner (1953) using crash data from rural three-leg intersections found that the frequency of collisions between vehicles turning around either shoulder was approximately proportional to the square root of the product of the traffic volumes on the main road and around either shoulder. Other early studies of intersection crashes (McDonald 1953; Raff 1953; Webb 1955) indicated that an increase in traffic on the major facility has a small effect on the crash rate, whereas an increase in traffic volume or an increase in the percent of traffic from the minor facility results in a rapid increase in the crash rate. Many studies have explored the relationship between crashes and descriptions of the features of intersections. For example, Hannah, Flynn, and Webb (1976) examined the relationship between crashes and characteristics of intersections in rural municipalities in Virginia. David and Norman (1975) analyzed the 3-year crash history of 558 intersections in northern Cal- ifornia. Categorical analysis methods were used, and results indicated that sight distance obstruction, street signs, use of left-turn storage lanes, use of raised marker delineation, pres- ence of bus loading zones, and multiphase signalization all affected crash rates. Still other studies concentrated on developing statistical models of the relationship between traffic crashes and geo- metric features. Bauer and Harwood (1996, 2000) developed statistical models incorporating the effect of traffic control features and traffic volumes on intersection crashes, and sta- tistical relationships between intersection crashes and charac- teristics (Bauer and Harwood 1996, 2000; Vogt 1999; Harwood et al. 2002; Washington et al. 2005) are the basis for the pre- dictive model for intersection crashes in the Interactive High- way Safety Design Module (IHSDM) (FHWA 2006). IHSDM is a suite of software analysis tools developed by the FHWA to evaluate safety and operational effects of geometric design decisions during the design process. Driver factors also influence crash risk. Information on drivers in state crash databases normally is limited; however, analyses have identified some patterns of driver-related fac- tors in intersection crashes. Of particular note is the con sistent finding that older drivers are overrepresented in intersec- tion crashes (e.g., Insurance Institute for Highway Safety 2007). Driver distraction has also been found to be a factor Safety Research Questions
9One specific point worth mentioning is the possible use of the system to track pedestrians and bicyclists. The Site Observer was not conceived for this task, but it has the potential to do so. As can be seen in Chapter 8, the imaging system tracks pedestrians, even though these tracks are rejected as noise in the current vehicle tracking implementation. Conflicts and Crashes at Intersections ⢠What are the differences in conflicts that result in crashes and those that do not? ⢠How do multiple drivers manage to resolve conflicts with- out crashing? ⢠What are the effects of volume, weather, and time of day on conflicts? ⢠Can distributions of conflict metrics at an intersection be directly related to safety and operation? ⢠What are the common factor dependencies of conflicts and crashes? Signalized Intersections Compliance with and Violations of Right-of-Way: Red Light Running ⢠Does the proportion of motorists violating red lights increase with the complexity of the intersection? ⢠What proportion of motorists violates the right-of-way rules by running a red light? ⢠What proportion of motorists who run red lights do so at the beginning of the red phase as compared to well into the red phase? This former indicates a deliberate action, whereas the latter probably is a nondeliberate action. ⢠How does the proportion of red light running affect crash risk? ⢠Does the proportion of motorists running red lights at an intersection vary with: 44 Approach speeds? 44 Approach traffic volumes? 44 Length of green phase of the signal? 44 Level of service of the approach? 44 Level of service of the intersection? ⢠Are there any changes in the signal phasing and timing that would reduce right-of-way violations at intersections with incidents of red light running? Signalized Intersections: Right Turns ⢠What is the distribution of accepted gaps for right turns with permitted right on red? How does this vary by road type and traffic volume? in intersection crashes; for example, Eby and Kostyniuk (2004) report that drivers in approximately 30% of intersection crashes were considered to be distracted or inattentive to the driving task. Surrogates for Intersection Crashes and Countermeasure Evaluation Although statistical models of intersection crashes may be determined from crash data, they can hardly address detailed questions of whether particular conflicts carry more or less risk or how specific countermeasures, such as changes in sig- nal timing, may affect driver behavior or crash risk. Relating crash risk to particular conflict types requires detailed crash and exposure data and detailed measurements of vehicle kinematics. In Chapter 5, intersection conflict measures are reviewed, but defined broadly, they are based on gap timing, range, and speed information that is not readily obtained using current technologies. The Site Observer opens up the possibility of capturing and evaluating conflict measures in great detail and thus relating their statistical patterns to those of actual crashes. This remains hypothetical, but once research can establish common patterns, metrics can be used as surrogates for crash. For example, to evaluate a particular countermeasure, before-and-after measures of conflict can be compared. What is crucial is that this can be expected to be achieved with sufficient statistical power in a few weeks or months, rather than after a wait of several years after crash numbers are reported. Arguably, such analysis may have immediate benefit; if the number and/or severity of traffic conflicts can be reduced using a countermeasure, it is reason- able to declare a safety benefit, even if the precise relationship to associated crash numbers is unknown. research Questions for Intersection Safety Many safety research questions are relevant to the Site Observer, although all are centered on conflict metrics and other measures of risk and safety; there is no expectation that the observer will be installed in sufficient locations or for suf- ficient time to allow detailed analysis of large numbers of crashes. However, metrics derived from vehicle trajectory data will allow in-depth analysis of conflicts and provide an opportunity to address the basic questions about the rela- tionship between conflicts and crashes. Below is a broad indicator of the kinds of research ques- tions that potentially can be addressed by the Site Observer. Other questions relate to unsignalized intersections, includ- ing roundabouts; Table 2.1 provides an additional summary, together with assessment of the likely applicability of the Site Observer.
10 Table 2.1. Summary of Research Questions for Intersection Safety Research Questions Data to Address the Questions Methods to Devise Countermeasures Evaluate CountermeasuresData from S09 Other Data Compliance/violations of right-of-way (signalized intersections): Red light running (RLR) ⢠What is the incidence of RLR? ⢠What are the crash risks of RLR? ⢠How can RLR be reduced? Vehicle trajectories, kinematics of conflicting vehicles Compare number and rates of incidents and crash risk across different road types, regions, traffic volumes, approach speed, levels of service 1. Change signal timing 2. Put in red light cam- eras (enforcement) Compliance/violations of right-of-way (signalized intersections): Right turns on red (RTOR) ⢠What is the incidence of RTOR when prohibited? ⢠What is the incidence of not yielding to pedestrians? ⢠What are the crash risks for RTOR compared with the benefits of increased throughput? Signal phasing and timing Compare number and rates of incidents and crash risk across different road types, regions, traffic volumes, approach speed, levels of service Calculate the decrease in delay from RTOR; compare mone- tary costs of crashes against costs of delay 1. Signage and signal for right lane 2. Enforcement cameras Costs and benefits of left-turn lanes and phases ⢠What are the crash risks of unprotected left-turn phases? ⢠How much do dedicated left-turn lanes decrease crash risk? Compare crash risks of unpro- tected left turns (and dedi- cated left-turn lanes) by road types, region, approach vol- umes, and levels of service Change signal phasing to protected left-turn phase Roundabouts ⢠What is the incidence of right-of-way violations in roundabouts? ⢠What are the crash risks of roundabouts? ⢠What is the effect of roundabouts on pedestrian crashes? ⢠What are the benefits of roundabouts in terms of safety and reduction of delay? Vehicle trajectories, kinematics of conflicting vehi- cles and/or pedestrians Calculation of delay, economic costs of delay Compare incidence of row vio- lation by region, approach volumes; calculate crash risk by region, approach volumes Compare to comparable crash risk and delay for intersections Gap acceptance (unsignalized intersection) ⢠What is the distribution of gaps for vehicles on minor legs to unsignal- ized intersections? ⢠What are the crash risks associated with those distributions? ⢠Are there ways to decrease the crash risks for minor road vehicles (for example, increase sight distance standards)? Calculate crash risk by approach volumes, region, and day-and-night conditions Evaluate increases in sight distance Access points near intersections ⢠What is the crash risk of access points within 200 feet of inter- sections? ⢠How far should the nearest access point be from an intersection so that it will not increase crash risk at the intersection? Vehicle trajectories, kinematics of conflicting vehi- cles or pedes- trians Compare crash risk across differ- ent road types, regions, traffic volumes, and approach speeds Evaluate closing off access points (continued on next page)
11 Table 2.1. Summary of Research Questions for Intersection Safety (continued) Research Questions Data to Address the Questions Methods to Devise Countermeasures Evaluate CountermeasuresData from S09 Other Data Compliance/violations of right-of-way (unsignalized intersections): stop and yield controls ⢠What proportion of vehicles on stop-controlled approaches do not stop? ⢠Does this vary by region or weather? ⢠What is the crash risk posed by this behavior? ⢠What proportion of vehicles on yield-controlled approaches do not yield? ⢠Does this vary by region or weather? ⢠What is the crash risk posed by this behavior? ⢠Can this behavior be reduced? Compare incidence of violations and crash risk across different road types, regions, traffic vol- umes, and approach speeds Evaluate use of enforce- ment cameras ⢠What proportion of drivers make right-on-red turns when such turns are prohibited? How does this vary by road type, sign location, and approach traffic volumes? ⢠How often is the right-of-way of a pedestrian violated by drivers making a right-on-red turn? What are the risks of pedestrian crashes from right turns at intersections? With permitted right on red? With prohibited right on red? By road type or traffic volume? Signalized Intersections: Left Turns ⢠What is the distribution of gap acceptance on approaches with unprotected left turns? How does it vary with volume, presence of a left-turn bay, and complexity of intersection? ⢠What is the crash risk of unprotected left-turn phases? ⢠What are the safety benefits of dedicated left-turn lanes? ⢠What are the safety benefits of dedicated left-turn lanes with protected left-turn phases? ⢠What is the proportion of drivers turning without making a stop? Access Points Near Intersections ⢠What is the effect on crash risk of commercial access points within 200 feet of intersections? ⢠How does crash risk vary with approach volumes and access point volumes? road Departure Safety Because road departure safety is one of the key research areas for the SHRP 2 Safety program, it is worth considering the extent to which the system can contribute to this research. Single-vehicle road departure crashes include crashes with roadside objects, rollovers, and collisions with other vehicles if the vehicle first ran off the road then reentered and collided with another vehicle. Selecting a site with known road depar- ture crash problems for a study may present technical, ethi- cal, and legal problems. From the technical standpoint, the question is one of camera placement; crashes and conflicts can be distributed over considerable distances, so where is the best location? On the other hand, if a site were selected because of a serious safety record and a crash occurred during the study, the road commission or state DOT probably would face legal suits for knowing that this was a dangerous location and not fixing it. Experimental methods associated with the deploy- ment of the Site Observer should take this into consideration, and selection of study sites most likely should be random from among typical examples of relevant road segments. The research team considered the kinds of research ques- tions that might be addressed via the Site Observer. The sys- tem may capture incidents of road departures, including events in which the vehicles almost ran off the road, ran off the road and stopped, ran off the road but recovered and either continued on the way or crashed into another vehicle, or ran off the road and crashed or overturned. Actual crash events will be rare, and even the events in which a vehicle runs off the road with no harm may not be sufficiently fre- quent to be captured in useful numbers. This again argues for the development and use of surrogates for running off the road events as an efficient and cost-effective way of studying the safety problem. In a recent SHRP 2 study (Gordon et al. forthcoming), it was hypothesized that vehicle road departure crashes occur only under conditions of disturbed control, for which dis- turbed control is an interruption or delay in the process of
12 Research Questions Data to Address the Questions Methods to Devise Countermeasures Evaluate CountermeasuresData from S09 Other Data ⢠Can episodes of disturbed control be identified from vehicle trajectory data? ⢠Can these episodes be used as surrogates for road departure events? Vehicle trajectories, kinematics of conflicting vehicles Vehicle- based data Check if there are identifi- ers in vehicle trajectories that correspond to met- rics of disturbed control from in-vehicle data If disturbed con- trol can be identified, it can be used in evaluations as a road depar- ture surrogate ⢠What is the relationship between road departure events and road departure crashes? Crash data for road segment Compare rate of road departure events (or sur- rogates) and crash rate on segment Effects of changing conditions at one site ⢠Effect of traffic on road departure events, crash risk? â Opposing lane, same lane? â Different mixes of vehicle types (trucks and passenger cars)? ⢠Effect of weather on road departure events, crash risk? ⢠Effect of light conditions on road departure crashes? ⢠Effect of work zones on road departure crashes? Vehicle trajec- tories, kine- matics of conflicting vehicles Monitor traffic volumes, weather, light con- ditions, and work zones Compare incidence of road departure events and crash risk between the levels of conditions Effects of roadway features across sites ⢠What is the effect of isolated horizontal curves on road departure events compared with that of a series of horizontal curves? ⢠What is the effect of shoulder width on road departure events? ⢠What is the effect of rumble strips on road departure events? ⢠What is the effect of edge line markings on road departure events? ⢠How does risk of road departure event vary with superelevation? ⢠How does the presence of spiral transition affect risk of road departure event? ⢠How does risk of road departure event vary with shoulder width, shoulder type? Compare incidence of road departure events, and crash risk between categories of roadway features ⢠How do changes in roadway features change incidence of road departure events and crash risk? ⢠What is the effect of each of the following: â Changes in lane width? â Introduction of rumble strip? â Introduction of rumble strip to center line? â Change in shoulder width? â Change in shoulder type? â Change in pavement markings? â Change in advisory signs? â Change in speed limit? Design experiments using before-and-after designs with controls or matched sites and conduct com- parison analysis Table 2.2. Summary of Research Questions for Road Departure Safety perception, recognition, judgment/decision, or action in the driving task. It was also hypothesized that crash surrogates for road departure crashes exist and are a combination of objec- tive measures of disturbed control and highway geometric factors and off-highway environmental factors. The question of development and validation of road depar- ture surrogates was addressed in that study; a number of candidate surrogates were proposed, and some level of valida- tion was carried out. One of the simpler candidate surrogates was time to edge crossing, the predicted time before a vehicle will leave the road (including passing across the shoulder) assuming it maintains its current path. In principle, the Site Observer can provide such surrogate information, although the accuracy in lateral position and velocity estimation is likely
13 to be very demanding. By comparison, an in-vehicle study that makes use of a lane tracker has more direct access to vehicle kinematics and driver actions (especially steering wheel move- ment). One area of application that is more likely to favor the site-based observer is for before-and-after study of counter- measures, such as when rumble strips are installed or lane marker clarity is improved. Conversely, when driver factors and disturbed control are concerned, it seems unlikely that the site-based system can compete with the in-vehicle record- ing approach. Table 2.2 summarizes a number of relevant research ques- tions for road departure crashes. It seems clear that the site- based system most naturally complements what is possible with an in-vehicle approach in conditions in which it is hard to capture highway and environmental factors from the vehi- cle or from spatial databases that can be linked to the driving data. For example, the effect of rumble strips may be esti- mated by their influence on a validated surrogate, especially if it can be seen that surrogates associated with drift onto the shoulder are significantly reduced. Overall, it seems plausible that the Site Observer can play a future role in analysis and countermeasure evaluation rele- vant to road departure crashes. However, for the current study, the focus is specifically on intersection safety problems.
