Compare object based classification of GSV images with remote sensing imagery
TODO: Insert image that summarizes repository
Author: Andrew Larkin
Co-Author: Xiang Gu
Affiliation: Oregon State University, College of Public Health and Human Sciences
Principal Investigator: Perry Hystad
Co-Investigator: Lizhong Chen
Date Created: November 14th, 2018
Summary
The purpose of this project is to evaluate whether GSV image estimates capture characteristics of the built environment, most notably characteristics dependent on visibility (e.g. beauty) compared to traditional estimates based on satellite imagery. The MIT Place Pulse dataset is the underlying dataset used for comparisons. Place Pulse contains georeferenced street view imagery and 1.5 million participant comparisons of images on characteristics such as beauty, safety, and liveliness. GSV images estimates were derived using a PSPNet model, while remote sensing estimates were derived using buffers around Pulse Place locations in ArcGIS. This work expands upon a previous project which compared satellite estimates of greenspace to percent green pixels in GSV imagery.
Repository Structure
Files are divided into two folders, based on their relative stage of project development. Scripts from a previous project were utilized for downloading remote sensing imagery and generating buffer based estimates
- BuiltEnvEstimates - Image and remote sensing estimates of the built environment (may not be available until publication).
- StatisticalAnalysis - Scripts to perform descriptive statistics and create regression models using built environment estimates.
External Links
TODO: insert link to publication once published
- MIT Place Pulse - http://pulse.media.mit.edu/about/
- Image Segmentation Analysis - TODO: insert Xiang Gu's github page
- Remote Sensing Buffer Estimates - https://github.com/larkinandy/LUR-NO2-Model
- Previous study comparing GSV percent green to remote sensing - https://github.com/larkinandy/GSV_NDVI_Comparison