determine the rate at which vehicles impact the concrete median barrier and estimate the proportion of all concrete barrier crashes that are reported to the police. A 1.14-mile section of eastbound Interstate 84 near Hartford was studied. The two-way traffic volume of the study section was approximately 130,000 vehicles per day, the average distance from inside travel lane to concrete median barrier was 2.5 feet, and the number of lanes varied from four to eight (both directions). A modular automatic road analysis (ARAN) system, essentially a modified van outfitted with various cameras and other onboard sensors and data collection equipment, was used for data collection. At the time of the study, Connecticut used two ARAN systems to assess the condition of all state-maintained roads each year. The researchers modified the ARAN so that a camera could be mounted to capture data from the left (e.g., median, side of the van). Similar to the Galati (1967) study, each barrier section was marked with an identification number and a baseline video of the study section was obtained. Then, five additional videos of the section were captured over the 8-month study period. Using dual videotape viewing equipment, a data collector manually compared each frame of the reference video with the corresponding frame of the subsequent video tapes and logged any new marks on the concrete barrier. A total of 62 collisions were documented. One limitation of the method noted by the authors was the exclusion of certain barrier sections, approximately 40 of the 302 available concrete barrier sections, due to drastic light changes (due to entering/exiting underpasses). Suggested improvements to the methodology included the use of a more permanent means of marking each barrier section, such as stenciling numbers with paint (instead of permanent marker) and whitewashing the barrier sections at the beginning of the data collection period (to eliminate the need for the reference video).

The primary strength of this method was the ability to estimate encroachment rate data for high-volume roadways with narrow medians and barriers present. Primary limitations of the method include no ability to collect trajectory information, no means of discerning vehicle type, and the barrier presence requirement limits the variation in site characteristics. For the Galati (1967) study, it was unknown what portion of reported w-beam damage observed was a result of mowing activities (in the portion of the section with turf median) and snow removal activities (during the winter months) and not unintentional vehicle encroachments. The Fitzpatrick et al. study did not suffer from this limitation, as the median on the study section was paved and data were not collected during the winter months. In both methods, the outlined procedure of a person manually reviewing films/images to identify and document barrier damage appeared labor intensive and time consuming (no specifics on labor hours were provided in either study). Modern image processing techniques, however, may be able to automate the detection process and effectively reduce this time requirement. Another limitation to both studies was that the data technically collected were crash rate data and would not detect encroachments that do not result in impact with the barrier. The narrow medians in both studies (5 feet or less distance from edge of travel lane to the barrier), however, should limit the number of vehicle encroachments that do not result in a barrier strike.

Maintenance/Repair-Based Methods

This method uses maintenance records to provide an estimate of encroachment rate. The method requires a longitudinal barrier to be present in the median or on the roadside and located sufficiently close to the travel way to limit encroachments that occur without impacting the barrier. Also, barriers that are unable to sustain multiple impacts without repair (e.g., flexible) are preferred. A single instance of this method was found in the literature.

Moskowitz and Schaefer (1961) investigated the performance of chain-link fencing and dual-sided w-beam barriers to serve as median barriers on California freeways. Test sections of both barriers

were installed on the Santa Ana Freeway (5.74 miles total; 3.17 miles of fencing) and Nimitz Freeway (6.74 miles total; 3.87 miles of fencing). Both roadways were six-lane freeways with 12-foot-wide medians, approximately level grade, and had traffic volumes between 90,000 and 100,000 vehicles per day. Crash data for 1 year before barrier installation and 1 year post barrier installation were reported along with corresponding maintenance information. Hutchinson and Kennedy (1966) used the Moskowitz and Schaefer reported fencing repair data to estimate median encroachment rate on both freeways. The underlying assumptions were that there were few, if any, vehicles able to recover in the 6-foot area on either side without damaging the fence and that even minor impacts to the fence would trigger a repair record (as opposed to minor w-beam barrier hits, which may not trigger repair).

The requisite conditions for this method (i.e., barrier present, close to roadway, and of a certain type) severely limit roadways that can be used for this method. In addition, the method provides only an estimate of encroachment rate and not full vehicle trajectory information. Although encroachments without barrier impact are assumed to be small in number given the site conditions, the estimated encroachment rate will exclude any encroachments that do not result in barrier impact.

Triggering Methods

Triggering methods use a device installed on the roadside or in the median that was able to automatically detect and record data pertinent to encroachments. The device was installed at one or more locations to collect encroachment data for that section over a specified time period. Typically, this method can provide encroachment frequency information and some details on the characteristics of the encroachment. Two instances of this method were found in the literature.

Hutchinson and Kennedy (1966) investigated using loop inductors to measure the encroachment time and speed of the encroaching vehicle. Small inductance coils (30" and 13" diameter, and 3 Ļ 6 feet) were installed in a concrete overlay on Route 47 just north of Gibson City and were generally successful. As a result, the researchers tested larger coils (5 Ļ 50 foot) in soil instead of pavement. The system functioned well during summer but not when temperatures were below freezing. No additional money or personnel were available, so work in this area was discontinued by the researchers. No further details of a ground-installed loop-detector type system were found in the more recent available literature.

Calcote et al. (1985) installed six GEM units at rural Texas locations between September 1979 and February 1980 (which remained in service until the end of 1980). Nearly 27,500 hours of operation resulted in capturing 31 unintentional events and 3,161 intentional events. At each GEM system installation, a camera unit was installed that was triggered to record when the GEM unit sensed an encroachment. Note that the researchers considered unintentional encroachments to include only totally out of control vehicles, vehicles leaving the roadway in a controlled but evasive manner, and vehicles that departed the roadway as a result of driver inattention. Six different sites totaling 3 miles included four 4-lane rural divided highway sections and two 2-lane rural highway sections. Details and a preliminary evaluation of the GEM device were reported by King and Cerwin (1981). The system uses the w-beam rail of the guardrail and one additional insulated conductor, suspended above the w-beam rail, to transmit one high- and one low-frequency (14.7 MHz and 147 kHz) electromagnetic wave. A built-in triggering system provides a record of up to 50 encroachments, and the entire GEM system can be operated continuously for approximately 1 week on a single 12-volt battery. Recorded data include the event time, the vehicle velocity both

parallel and perpendicular to the guardrail, the angle of the vehicle, location of the vehicle along the guardrail length, the vehicle size, and a plot of the vehicle path. Several non-impact tests were conducted with varying vehicle sizes (Volkswagen Beetle through tractor-trailer combination) to determine the accuracy of the developed system. The tests indicated vehicle speed was +/- 0.5 mph, angle measurement was +/- 1 degree, and vehicle location along the guardrail was +/- 10 feet. The parallel and perpendicular ranges of the system were found to be greater than 1,300 feet and 3 feet, respectively. The authors also noted a variation of the GEM system that could be installed in a location without guardrail present. Other than the Calcote et al. (1985) study mentioned above, no more recent uses of the GEM were found in the published literature.

Calcote et al. (1985) did mention considering a radar-based system, the Encroachment Recorder, developed by Cromack Engineering Associates of Tempe, Arizona. The system was eliminated from consideration as the company had not yet developed a prototype system that could be evaluated or used in the study. No further details of the system were reported by Calcote et al. (1985), and no other information regarding this type of system was found in the more recent literature, suggesting that the system may have never been fully developed. There was a radar-based system found in the literature that provides automatic detection of large animals near roadways and warns nearby motorists via a flashing sign (Huijser et al. 2009). It may be possible to modify this animal detection system to provide encroachment detection capabilities, but it appears no radar-based method specifically designed for encroachment detection exists.

The primary advantage of these triggering methods was the automatic detection and recording of encroachment events. The inductance method would allow collection of time and vehicle speed while the more sophisticated GEM system allows collection of speed, angle, and location, that is, essentially the same information collected by previous manual field collection methods. A GEM typically requires guardrail presence, which would limit the roadways on which it could be installed. Newer technology may reduce the cost (there was no cost mentioned in the report) and/or increase the capability of the GEM. The primary limitation of the system would be the cost to physically install and monitor either system. The cost would likely limit the total number of installations, effectively limiting the variation in roadway characteristics over which encroachment data could be collected. Other GEM system limitations noted by Calcote et al. (1985) were issues with false triggering, loss of data in cases where the barrier was impacted, and a nominal recording time of 4 seconds per event, which was not sufficient to capture shallow, low-speed encroachments.

Site Specific Video-Based Field Collection Methods

This method involves installation of a video camera at one or more locations to provide continuous video footage of the travel way edges of a roadway section. Researchers then examine the collected video footage to detect vehicle encroachments onto the median and/or roadside. For each encroachment, the video can be used to determine vehicle type and estimate relevant encroachment characteristics such as speed, angle, and maximum lateral extent. This same overall process was repeated for different collection sites (e.g., roadway segments). A single instance of this method was found in the available literature.

Calcote et al. (1985) installed four video tape recorders on urban Texas highways from September 1979 to July 1980 and again from September 1980 to January 1981. Researchers examined compressed versions of the video recordings for evidence of both intentional and unintentional encroachments at each of the four sites. Once an encroachment was detected, the available video

data were used to determine vehicle size and estimate encroachment characteristics such as departure speed, encroachment angle, maximum lateral encroachment distance, and longitudinal distance traveled. Just over 36,000 hours of video resulted in 12 unintentional encroachments and 6,358 intentional encroachments. Again, the same definition of unintentional encroachment was used as mentioned in the triggering method section above. Using this definition, the researchers subjectively classified encroachments as intentional or unintentional based on the video of each collected encroachment event. The authors noted that only video of the encroachment provided an adequate, albeit subjective, method of distinguishing intentional from unintentional encroachments. In total, seven different urban locations with a total of 1.82 miles of roadway were captured. The traffic volumes ranged from 17,863 to 92,314 vehicles per day. Due to the low number of unintentional encroachments initially detected, the researchers experimented with different camera locations in hopes of increasing the coverage range of each camera. Higher mounting heights resulted in greater ranges, but at the expense of an ability to distinguish encroachments and adequately measure encroachment characteristics. The practical coverage limit for each camera was found to be approximately 0.25 miles. Other issues noted by the researchers included availability of continuous power at potential mounting locations and equipment malfunctioning.

Primary advantages of this method include producing a visual record of the observed encroachments that can be used to determine encroachment frequency and relevant encroachment characteristics. The method also allows the possibility to distinguish between intentional and unintentional encroachments. The primary limitation of this method was the line-of-sight limitations of the camera equipment, which increases the cost and time required to collect a substantial amount of encroachments across varying roadway and traffic conditions. Other limitations include finding a site suitable for camera mounting (e.g., height and continuous power available), and roadway section lighting was required to allow for usable nighttime recording.