Dynamic passive pedestrian detection (DPPD) has the potential to enhance the operational efficiency and safety of signalized intersections by cancelling unnecessary pedestrian service calls or extending the pedestrian phase to allow a pedestrian to safely finish crossing the street. Researchers examined the accuracy and reliability of two thermal sensors (high and low resolution) and one optical sensor for DPPD at one signalized intersection and one signalized mid-block trail crossing in Washington County, Oregon in the Spring and Fall of 2019. The signalized intersection had crosswalk lengths of 120 feet and 90 feet, and the mid-block location had a street-crossing width of about 70 feet, comprised to two crossings of 30 feet separated by a median. Due to crossing length, two curb zones and two or three crosswalk zones were defined for each crossing in the detection software for individual evaluation, and zones were classified as arriving or departing based on the pedestrian’s travel direction. Video files recorded between March and May 2020 were analyzed to assess the accuracy of the sensors according to the following definitions:
Valid – Pedestrian enters zone, detection is initiated and held continuously until pedestrian leaves, with less than one second of delay in initiating or dropping the call.
Inaccurate – Any of the following detection outcomes were counted as inaccurate:
- Spotty – Pedestrian in the zone while call is dropped and reinstated.
- Dropped – Pedestrian in the zone while call is dropped and not reinstated.
- Held – Pedestrian leaves the zone, but call is held by more than one second.
- Late – Pedestrian enters the zone, but call is initiated late by at least one second.
- Miss – Pedestrian enters and leaves the zone without detection.
Accuracy (percent) = 100 x Valid / (Valid + Inaccurate)
A total of 512 unique crossing observations were recorded, 102 at the signalized intersection and 410 at the mid-block trail crossing.
- Consider the specific use when selecting a sensor. Examine the sensor specification ratings, sensitivities to various environmental conditions, desired use case functionality, and agency goals in evaluating different sensors before making a selection. For example, in this study thermal sensors performed better under low light and adverse weather conditions than sunny and clear conditions, and had a higher overall accuracy rate compared to that of the optical sensor. However, optical sensors may be more advantageous when an agency desires to simultaneously use the sensor’s video feed to monitor an intersection.
- Carefully plan the sensor selection process. This process may include a site‐visit and/or coordination with the sensor vendor’s technical support. Significant installation and calibration efforts were included by the sensor vendors for this study. It is important to get the camera make/model, mounting location, view orientation, and mounting height correct to facilitate accurate detection.
- Consider site characteristics when planning detection zones. Local site characteristics like overhead utility lines, trees or other vegetation, span wire versus mast arm signal head mounting, and signage can all affect the line‐of‐sight for potential detection zones.
- Remember to determine the required detection distance and angle of a sensor system. These must be considered in tandem with site characteristics because conflicts with the existing infrastructure can induce false detections or occlusions.
- Account for site-specific conditions during calibration and placement of sensor detection zones. Detection zones need to be large enough to capture crossing pedestrians but compact enough to exclude others who do not intend to cross. Distance and location can affect the required horizontal detection zone sizes.
- Select a sensor system suitable for the desired functionality of DPPD. If pedestrian call extension functionality is desired, the detection system will need to accurately track a pedestrian for the entire crossing movement, and one or more detection zones must monitor the entire length of a crosswalk. The intersection size will inform sensor selection.
Volume 8, November 2020