Evaluation of dynamic passive pedestrian detection

Abstract

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. This paper examines the accuracy and reliability of two thermal sensors and one optical sensor for DPPD at one mid-block crossing location and one signalized intersection in Washington County, Oregon. Results indicate the average accuracy rate for the thermal sensors was 89% at the signalized intersection and 82% at the mid-block location. The most common inaccurate detection types for the thermal sensors were Late, Held, and Miss calls. Average accuracy rate for the optical sensor was 26% at the signalized intersection and 83% at the mid-block location. The most common error types for the optical sensors were Spotty, Late, and Miss calls. False detections most often occurred for both sensors when vehicles and bicycles in the roadway crossed into the detection zones. Weather and lighting conditions affected thermal sensor accuracy, while only pedestrian type affected optical sensor accuracy. Pedestrian curbside waiting habits, jaywalking, and implications for sensor selection, deployment, and development are discussed.