Crowdsourced street view imagery and urban informatics

Keywords: street-level imagery, computer vision, VGI, morphology
Abstract Type: Paper Abstract

Abstract

Street view imagery (SVI) has been used extensively in the past decade. Numerous research papers across a variety of disciplines have been using this emerging form of data to infer the perception of streetscapes, understand environmental conditions, map street furniture, and measure urban form elements such as the green view index. However, nearly all such studies have relied on high-quality panoramic imagery sourced from commercial services such as Google Street View and Baidu Maps. In parallel, crowdsourced SVI, a sequential visual variant of VGI, has been emerging, thanks to platforms such as Mapillary and KartaView and myriads of volunteers collecting data from smartphones and dashcams. While millions of images are released on a daily basis for free use around the world, their use in urban informatics is virtually non-existent. Previous studies have revealed quality issues such as non-panoramic imagery and heterogeneous completeness, making them difficult to compete with readily available commercial counterparts covering most of the globe. This presentation will reveal the usability of such data, primarily in inferring the urban form at a large-scale. Preliminary results indicate that volunteered SVI, in most cases, may derive accuracy that is on a par with commercial SVI, and has some advantages such as mapping urban areas not yet covered by commercial services.