Pedestrian Volumes from Push-Button Traffic Signal Data in Oregon: Estimating Models and Assessing Model Transferability

This study’s objective was to develop methods to estimate pedestrian volumes from pedestrian push-button data and to assess the transferability of those methods. First, videos were recorded in Summer 2022 at 64 signalized intersections in Oregon, and pedestrian counts were extracted. Next, pedestrian traffic signal data were assembled for the same locations and time periods. After counting 41,843 crosswalk users from 2,679 h of video, the final sample included 8,546 observations, each representing 1 h at one signalized crosswalk. Data analysis utilized a structured multistep process. Three dependent variables represented different interpretations of “pedestrians” or “crosswalk users.” Eight independent variables were tested, derived from pedestrian actuations or detections and adjusted to reflect signal operations and push-button press behaviors. Five functional forms (e.g., linear, quadratic, etc.) were compared, and 14 segmentation variables were considered, which captured signal operations, temporal factors, and built-environment characteristics. The final recommended model had a quadratic relationship between total crosswalk users and the number of push-button presses, filtered to remove multiple presses within 15 s of another. The model’s predictions were strongly correlated (0.86) with observed pedestrian volumes and had a low average error of ±2.4 pedestrians per hour. Furthermore, when comparing the Oregon and Utah models and data, the remarkable consistency of the two states’ quadratic fit lines—collected in different states 3 years apart—gives some expectation of transferability for the developed model. Such pedestrian volume estimates could improve pedestrian traffic monitoring, safety assessments, and equity and health analyses.