'Big data' for pedestrian volume: Exploring the use of Google Street View images for pedestrian counts. (September 2015)

• Record Type: Journal Article
• Authors: Yin, Li; Cheng, Qimin; Wang, Zhenxin; Shao, Zhenfeng
• Is Part Of: Applied geography. Volume 63(2015:Sep.)
• Page Start: 337
• Page End: 345
• Publication Date: 2015-09
• Subjects: 'Big data' -- Google Street View -- Pedestrian count -- Walkability; Geography -- Periodicals; Human geography -- Periodicals; Human ecology -- Periodicals; 910
• DOI: 10.1016/j.apgeog.2015.07.010
• Languages: English
• ISSNs: 0143-6228

Abstract

Abstract: New sources of data such as 'big data' and computational analytics have stimulated innovative pedestrian oriented research. Current studies, however, are still limited and subjective with regard to the use of Google Street View and other online sources for environment audits or pedestrian counts because of the manual information extraction and compilation, especially for large areas. This study aims to provide future research an alternative method to conduct large scale data collection more consistently and objectively on pedestrian counts and possibly for environment audits and stimulate discussion of the use of 'big data' and recent computational advances for planning and design. We explore and report information needed to automatically download and assemble Google Street View images, as well as other image parameters for a wide range of analysis and visualization, and explore extracting pedestrian count data based on these images using machine vision and learning technology. The reliability tests results based on pedestrian information collected from over 200 street segments in Buffalo, NY, Washington, D.C., and Boston, MA respectively suggested that the image detection method used in this study are capable of determining the presence of pedestrian with a reasonable level of accuracy. The limitation and potential improvement of the proposed method is also discussed. Highlights: Google Street View provides street images in seven different resolution levels. Level 3 images should be used for automatic pedestrian detection, if the default ACF training set and parameters are used. A re-training is needed if level 4 or higher images are to be used to match the parameters of ACF and the Google images. The image detection method proposed is capable of determining the presence of pedestrian with a reasonable accuracy level. This method can help to get an objective and large scale reliable estimate of pedestrian volume.