Improve the Validity of V2V Alerts with Better Accounting of Relative Elevation and Heading Between the Host Vehicle and Remote Vehicle, found by THEA Connected Vehicle Pilot.
USDOT Assessed the Safety Impact of THEA Connected Vehicle Pilot and Found Communication Between Vehicles Was Successful, but the Results of the Deployment Were Indeterminate due to a High Percentage of Invalid Alerts.
Date Posted

Safety Impact Assessment of THEA Connected Vehicle Pilot Safety Applications

Summary Information

The Tampa Hillsborough Expressway Authority (THEA) worked with the United States Department of Transportation (USDOT) to implement real-time Vehicle-to-Infrastructure (V2I), and Vehicle-to-Vehicle (V2V) communications under the Connected Vehicle Pilot Deployment Program. This study presented the technical approach, data analysis, and results of the independent evaluation of the safety impact of V2V and V2I applications deployed in a real-world environment on public roadways in downtown Tampa, Florida. For the study, THEA solicited volunteer drivers who frequently travel in the deployment area during their daily commute into downtown Tampa. Over 800 privately owned vehicles of volunteers and seven fixed guideway trolleys were equipped with aftermarket CV devices that could issue visual warnings to the vehicle operators.

A total of four V2V and three V2I applications were implemented. The V2V applications included forward crash warning (FCW), emergency electronic brake light (EEBL), intersection movement assist (IMA), and vehicle turning right in front of transit vehicle (VTRFTV) warning. V2I applications included pedestrian collision warning (PCW), end of ramp deceleration warning (ERDW), and wrong-way entry (WWE) warning. The V2V and V2I safety applications operated in two different modes during the deployment: (1) Silent mode, where the applications were operating in the background but did not issue any alerts to the drivers, and (2) Active mode, where the applications were fully active, issuing visual feedback to the drivers.

Data was collected from the vehicles' onboard devices over a period of 16 months, from May 2019 - June 2020.

Lessons Learned

Some key lessons learned from the Tampa, Florida CVP program included:

  • Utilize secure data storage platforms when analyzing data that may contain personally identifiable information (PII). Due to the existence of PII in THEA’s records, a secure data storage and analysis platform was required to protect study participants’ privacy. Data log files from the study were uploaded on a nightly basis to U.S. DOT’s Secure Data Commons (SDC) platform to support analysis by the independent evaluators and other U.S. DOT partners. The SDC platform provides data storage and processing functionality, as well as controlled access to these datasets for analysts from U.S. DOT managed, cloud based desktops.
  • Match silent and active alert samples by alert-specific kinematic conditions. Silent and active alert events should be assembled by similar initial kinematic conditions at alert onset in order to compare HV driver response between the two alert modes under the same conditions for each safety application. The initial conditions considered for statistical matching will depend on the alert type being studied. For example, for FCW alerts, initial conditions used for matching include time headway, host vehicle speed, and range rate between the host and remote vehicles at alert onset. 
  • Account for difference in elevation and heading between the host vehicle and remote vehicle for IMA and FCW applications, and adjustments to the timing of WWE alert applications. It was stated in this study that the valid alerts were infrequent, and changes needed to be made to the alert system. Only 33 percent of FCW alerts, 21 percent of IMA alerts, and 20 percent of VTRFTV alerts were found to be valid alerts. Adjustments were needed to be made to adjust for timing and differences in elevation in order to increase the percentage of valid alert.
  • Supplement analysis with tools that allow you to visualize event data. The team developed an event visualization tool that animated the motion of HV and RV using their instantaneous locations before and after an alert event, overlaid on a satellite image map. This allowed for visualizations of vehicle locations, movements, and interactions during an alert event. This visualization feature became instrumental in categorizing the alert event data as valid or invalid alert events, and allowed the team to gain a better understanding of the HV behavior during vehicle interactions.

Keywords Taxonomy: