Bridging the gap between pedestrian and street views for human-centric environment measurement: A GIS-based 3D virtual environment

Street View Image (SVI) data has gained popularity as a methodological tool for assessing pedestrian-level environments across various disciplines. Several studies in transportation and public health fields posit that environmental features measured through SVIs are similar to those perceived by pedestrians, whereas, in fact, SVIs are typically collected from vehicles. As a result, the SVI data may reflect significant differences from an actual pedestrian's view. Hence, this study investigates the errors in measuring neighborhood environmental features resulting from the difference between pedestrian and vehicle views. To accomplish this goal, a three-dimensional (3D) urban model is developed, utilizing multiple Geographic Information Systems (GIS) data. This model accurately represents the spatial positioning and dimensions of various entities. From the 3D model, virtual SVIs for each perspective are collected. Subsequently, we quantify environmental features in a manner similar to semantic segmentation for the images. Our findings reveal significant measurement errors for sidewalks, greenery, and roads between pedestrian and vehicle views. These errors are more pronounced with certain environment characteristics and parameter settings for image acquisition. For example, narrow road segments with few on-street parked vehicles exhibit increased measurement errors for greenery. Overall, the study provides guidelines for researchers on utilizing SVI to achieve reliable measurement of environmental features and move toward human-centric measurements.