Project Name

Infrastructure Safety Assessment Using Connected Vehicle Data

Research Team

Brian L. Smith, PE

Robert Kluger

Hyungjun Park

Sponsor

Connected Vehicles/Infrastructure University Transportation Center: Connected Vehicles/Infrastructure University Transportation Center

Project Dates

September 2012 - 2015

Project Description

The objective of this project is to investigate the feasibility of using connected vehicle data to create a system to identify infrastructure safety problems based on vehicular data indicating a high frequency of evasive driving maneuvers indicative of near-misses. Specific objectives of this study can be summarized as follows;

1. To identify thresholds for vehicle maneuvers (i.e. lateral control and accelerations) that may indicate an evasive maneuver.
2 To delve into statistical tests to combine frequency of maneuvers with location data to identify hot spots.
3. To demonstrate and test this approach using the UTC Northern Virginia testbed.
4. To validate this approach by comparing results to traditional crash data analysis results. 

The final success of this project will be determined by the ability to identify hot spots more quickly and accurately than using traditional methods. Considering that near misses are much more frequent than collisions, there should be evidence of a hot spot more quickly because of the ability to observe a near miss, combined with the accurate location of the vehicles involved using connected vehicles. Final results of this project will include the development of technique for network screening using connected vehicle technology.

Photos and Media

UTC Report Infrastrucure Safety Assessment

Methodology

The following tasks will be completed in order to meet the objectives of this project.
1. Analyze vehicular data available from connected vehicles to identify measures that may be used to identify evasive maneuvers.
2. Develop threshold values to use in extracting evasive maneuvers from vehicular data.
3. Use network kernel density algorithms to identify hot spots from vehicular data.
4. Apply results of tasks 1-3 to demonstrate and test the system on the UTC Northern Virginia testbed. 
5. Compare hot spots identified in Task 4 with traditional crash data analysis to determine level of correlation.
Task 1 – Task 1 consists of analyzing literature and collecting background information. This includes, but is not limited to, an in depth look at the connected vehicle technology currently available in the UTC Virginia testbeds, a detailed look at frequent evasive maneuvers taken during a roadway incident, as well as analyzing available vehicular data. The available vehicular data from the following sources will be used to hypothesize which variables (speed, deceleration, application of brakes, etc.) are most important in defining a near miss incident or use of evasive maneuvers: 
1.Strategic Highway Research Program II (SHRP II) conducted by Virginia Tech Transportation 
Institute (VTTI)
2.100-Car Naturalistic Driving Study conducted by VTTI
3.Other connected vehicle testbed data
Using the information from these sources, variables that may be able to be used as safety performance measures will be further analyzed in task 2 of this assessment. The J2735 standard message set includes numerous probe data elements that can be used in this study. This includes the following variables among other variables that may be less relevant in evasive maneuvers:
1. Acceleration
2. Brake Applied Pressure
3. Brake Boost Applied
4. Coefficient of Friction
5. Driving Wheel Angle
6. Latitude
7. Longitude
8. Obstacle Direction
9. Obstacle Distance
10. Speed
11. Stability Control Status
12. Steering Wheel Angle
13. Steering Wheel Rate of Change
14. Traction Control State 15.Yaw Rate 

Task 2 – Task 2 involves determining threshold values that define a near miss incident or extreme evasive maneuvers using the performance measures selected in task 1. By applying the thresholds to the SHRP II and other testbed data, a list of every location where a near miss incident occurred will be acquired. A combination of network kernel density (KDE) analysis as well as Getis-Ord (GI*) analysis will be used to determine the probability of any location being a hot spot. Results will be compared to existing crash data for validation.

Task 3 – In Task 3, the research team will develop and prototype a safety screening application to operate using data collected by the standard Virginia Tech Transportation Institute data acquisition system on the UTC Northern Virginia testbed. This will allow for the demonstration and testing of this concept. Note that this approach allows for a very low-risk approach to field testing the concept. For the sake of the test, it will not be necessary to transmit data from the on-board acquisition system using the DSRC infrastructure given that this is not a “real-time” application. The research team will periodically acquire the vehicular data, apply the thresholds identified in Task 2, and automatically determine if there is a certain area that frequently receives data that exceeds the thresholds determined for chosen performance measures. Near miss and evasive maneuvering data will be compiled and evaluated using a KDE and GI* techniques in order to identify hot spot locations along the UTC Northern Virginia testbed. Kuo et. al. showed that using both was important because they will indicate different areas for secondary hot spots in the results. This is because GI* is based on the number of data points while network kernel density estimation detects points with extremely high values. Both are important in locating hot spots. This phase will take the most amount of time, however preparation for task 4 may be started while the system collects the data needed. 
Task 4 –In Task 4, possible hot spot locations identified using the prototype system from Task 3 will be compared to traditional crash data (obtained from the Virginia Department of Transportation) to look for correlation. If traditional data indicates hot spots at the same point as the connected vehicle data collected indicates them, that there is most likely a correlation. There should be at least the same number, or more hot spots indicated as traditional data indicates since the idea of using connected vehicle data is to find at risk areas more quickly than waiting for incidents to occur and be reported. Once hot spots are determined, the location’s physical features will be carefully examined to determine if the hot spot is indicated due to random occurrence or some flaw (e.g. sight distance violation) with the design of the infrastructure. 
Task 5 – In the final task, a plan for widespread integration of this process into connected vehicle roll out will be developed. Findings could be outlined extensively in the Highway Safety Manual’s Roadway Safety Management Program, primarily in the network screening and safety effectiveness evaluation steps. State DOT’s and private traffic data collection companies could also use these concepts once connected vehicles become standard practice. Additionally, there are plans for future testing and analysis depending on the results of this assessment. Success of this project will be determined by the ability to identify hot spots more quickly and accurately than traditional ways as well as progression of connected vehicle technology.

Findings

Poster