Methodology for applying a crowdsourced street-level imagery data to evaluate street-level greenness

Keywords: street-level greenness, crowdsourcing street-level imagery, Mapillary, image filtering
Session Type: Virtual Paper Abstract
Day: Saturday
Session Start / End Time: 2/26/2022 02:00 PM (Eastern Time (US & Canada)) - 2/26/2022 03:20 PM (Eastern Time (US & Canada))
Room: Virtual 66

Abstract

Street-level greenness visibility has been reported to be associated with various health benefits but is difficult to quantify. Recently, a novel method using street-level mapping data (typically from Google Street View (GSV)) provides an opportunity for obtaining the area-level greenness value in an efficient way. Nevertheless, the latest GSV guidelines indicated that using applications to analyze and extract information from the Street View imagery is prohibited. Finding alternative mapping data is crucial for future research on this topic. Recently, crowdsourcing has become a popular approach to create open data at large-scale and in a shorter time. Notably, Mapillary is widely known as a crowdsourced street-level imagery platform, assessing street-level imagery data from all over the world. 　
This study represents a methodology for applying street-level mapping data from Mapilalry to evaluate street-level greenness. Since some of the street-level images on the Mapillary platform are not qualified for assessing street-level greenery, the focus of this study is on a framework for filtering images. Mapillary images are analyzed and the usable images are identified with a trained XGBoost classifier.
Results indicated that the street-level greenness value based on the filtered Mapillary data is closer to the results of the audit survey by the government, compared to the original Mapillary data. We suggest the capacity to assess street-level greenery and the potential of guiding urban landscape planning on crowdsourced street-level imagery.