Use of Thermal Imaging Cameras in NYC for Pedestrian Detection Had the Highest Accuracy Rate (87 Percent) at Night, Compared to Other Times of Day.
The Connected Vehicle Pilot Deployment (CVPD) Program in NYC analyzed the results of a pedestrian detection deployment using thermal imaging cameras.
New York City, New York, United States
Connected Vehicle Pilot Deployment Program Phase 3, Understanding and Enabling Cooperative Driving for Advanced Connected Vehicles in New York City – New York City Department of Transportation (NYCDOT)
Summary Information
Cooperative Driving (CD) is a means of sharing information to support vehicle movements to improve traffic safety and traffic flow. Vehicle-to-Everything (V2X) communication services enable the synchronization of vehicles via information exchange to coordinate, validate, and support drivers’ actions. As a part of the USDOT Connected Vehicle Pilot Deployment (CVPD), this project explored the applicability of cooperative driving for advanced connected vehicles (CD for ACV) on urban roadways and identified three testable use cases for New York City (NYC):
- Pedestrian and bicyclist safety through cooperation
- Cooperative work zones
- Cooperative intersection management.
This project identified data needs and assessed the existing and potential data sources that may contribute to CD for ACV. Specifically, Forward Looking Infrared (FLIR) thermal-based pedestrian detection technology that was used for the Pedestrian in Crosswalk application in the CVPD was evaluated.
For the Pilot, 10 intersections were outfitted with FLIR thermal imaging cameras capable of detecting pedestrian presence, pedestrian counts, and detection latency for in-time delay (delay between the time when a pedestrian walks into the detection zone and the time when the pedestrian is detected by the pedestrian detection equipment) and out-time delay (delay between the time when a pedestrian walks out of the detection zone and the time when the pedestrian is no longer detected by the pedestrian detection equipment.) The cameras were placed at both high and low pedestrian density intersections. Over 20 hours of detection video data were collected at two instrumented intersections from August 16 - August 21, 2021 from various scenarios involving different days of week, times of day, traffic demands, and light conditions.
Accuracy, precision, recall, F-1 score, and false negative rate were used as performance metrics for pedestrian presence detection and pedestrian counts.
FINDINGS
- The results showed that mid-afternoon and night-time had the highest accuracy rates in terms of pedestrian detection for both high and low pedestrian density intersections using the thermal cameras.
- The results at the high-density pedestrian site showed a 0.87 accuracy rate at night, compared to 0.81 in the morning, 0.83 in the mid-afternoon, and 0.68 at the PM peak time. Nighttime detection also had the highest recall, F1 score, and the lowest false negative rate compared to other times of the day at this site.
- The results at the low-density pedestrian site showed a 0.9 accuracy rate at night, compared to 0.57 in the morning, 0.85 in the mid-afternoon, and 0.63 at the PM peak time. Nighttime detection also had the highest recall, F1 score, and the lowest false negative rate compared to other times of the day at this site.
