A novel approach for estimating sidewalk width from street view images and computer vision

Sidewalks facilitate pedestrian movement and improve urban accessibility by providing safe and convenient pathways exclusively for pedestrians. While comprehensive sidewalk inventory data, including location and width, are essential for planning pedestrian infrastructure, traditional data collection methods are often costly and resource-intensive. Although recent advancements in deep learning enable the extraction of sidewalks from aerial imagery, these methods have several limitations. They require high-resolution orthorectified imagery and standardized training data, which are not universally available and are limited by occlusions such as dense tree canopies. To address these limitations, we propose a novel methodology for estimating sidewalk width using globally accessible street view images. Our approach utilizes computer vision to extract sidewalk features from images captured at two pitch angles, applies Canny edge detection to identify sidewalk edges, and uses trigonometric functions to calculate sidewalk width. With minimal input requirements, this method offers a cost-effective alternative for regions lacking high-resolution aerial imagery or resources for traditional data collection methods. Experimental validation demonstrates the method’s reliability, while error analysis identifies limitations and provides insights for future improvements. By offering a scalable and accessible tool, this approach supports the creation of comprehensive sidewalk inventory data, laying a foundation for informed and efficient urban design.