Deployment of Thermal Pedestrian Detection Technology in Tallahassee and Tampa, Florida Was Evaluated for Accuracy in Detecting the Presence and Movement of Pedestrians.
Tallahassee, Florida, United States
Tampa, Florida, United States
Integration of a Robust Automated Pedestrian Detection System for Signalized Intersections
Summary Information
Infrastructure to warn drivers of pedestrians in crosswalks typically relies on pedestrians pressing an activation button, yet in many cases this does not occur. Researchers investigated three automatic pedestrian detection systems capable of activation without needing a manual button press. These systems can be used in safety enhancement scenarios such as at midblock crosswalks to activate rectangular rapid flash beacons (RRFB), and at signalized intersections to place a call to the traffic signal controller for a pedestrian crossing phase. Three automatic pedestrian detection systems, two with microwave radar technology and one with thermal machine vision technology, were selected for testing based on an initial review of limitations and availability. The three selected systems were first tested under controlled conditions in the laboratory to evaluate functionality and determine whether they met the basic requirements for initial field testing. The systems were then tested at a midblock crosswalk and a signalized intersection inside a university campus without being connected to traffic control devices or warning lights to enable a concurrent comparison of detection performance. The research team established definitions for evaluation criteria as follows:
- Detection (yes/no) – was system able to detect pedestrian walking inside detection zone?
- Direction Detection (yes/no) – was system able to detect pedestrian walking in correct direction?
- Presence (yes/no) – was system able to detect pedestrian as long as pedestrian was inside detection zone?
Measures used in the testing included:
- Detection System Accuracy (Percentage) = Number of True Detections / Total Number of Detections
- False Detections (Percentage) = Number of False Detections / Total Number of Detections
- True Detection Accuracy (Percentage) = Number of True Detections / Total Number of Pedestrian Presence
- Percent of Missing Detections = Number of Missing Detections / Total Number of Pedestrian Presence
Tests were conducted to examine how well the systems detected pedestrians passing in a detection zone under different movement scenarios, wearing different clothing, and under daytime or nighttime lighting conditions. The system with the best test results in the initial comparison, the thermal machine vision system, was then deployed at three locations for further evaluation in Tallahassee and Tampa:
- Two midblock locations connected to RRFBs
- One signalized intersection connected to the traffic signal controller
- To enable assessment of detection accuracy, researchers installed video cameras to record observations for later comparison with the system’s detection output.
- All three systems were able to effectively detect pedestrians and place a call for an RRFB or pedestrian signal. However, the thermal system was the only one of the three systems able to recognize when pedestrians were moving away from crosswalks and to maintain detection regardless of pedestrian movement inside the detection zone, even if the pedestrian remained relatively still.
- The deployed thermal system was able to detect pedestrians approaching midblock crosswalks with an average of 92 percent detection system accuracy, and 2 percent false detections.
- The deployed thermal system at the signalized intersection was able to detect a waiting pedestrian with an average detection system accuracy of 94 percent, and was able to place a pedestrian service call 90 percent of the time.
- Testing showed that the thermal system was able to detect cases where the pedestrian left the detection zone early in 98 percent of cases, and resulted in not placing or removing the call to the signal controller 97 percent of the time when this occurred. This capability, to administer the removal of a pedestrian call when no longer needed, is useful for minimizing unnecessary vehicle delay.
