Skip to content

Reduced Version of GSV2SVF to be incorporated into wrapper program

Notifications You must be signed in to change notification settings

NewSoupVi/GSV2SVF-reduced

 
 

Repository files navigation

This is a modified version of GSV2SVF that can be started with a .bat script taking a task folder as a launch argument, and also looks for predownloaded images in that task folder. It is used by SkyViewFactorTool, a program that will automatically fetch GSV panoramas around a point and then automatically run GSV2SVF on all of them to extract their Sky View Factors and average them.

Original Readme

1. Background

GSV2SVF is designed to interactively calculate sky/tree/building view factors from Google Street View GSV) panoramas in Google Maps. In GSV2SVF, a GSV panorama is first classified at the pixel level into multiple classes including sky, tree and building, and then transformed into a hemispherical (fisheye) image. The sky, tree and building view factor respectively represents the areal proportion (in the range of [0,1]) of a fisheye image that is occupied by sky, tree and building pixels. A video is available at https://github.com/jian9695/GSV2SVF/blob/master/Video.mp4 and https://youtu.be/k00wCnuzuvE

2. Software infrastructure and system requirements

The Caffe-SegNet deep convolutional framework is used to classify street images (https://github.com/alexgkendall/caffe-segnet). Python and JavaScript were used to develop the interactive functionality that integrates Caffe-SegNet with Google Maps. The program currently runs only on Windows due to the restriction that the Caffe-SegNet module was compiled on Windows, although it could also be complied on Linux. A NVIDIA graphics card that supports CUDA 7.5 or newer versions is required. It has been tested only on Windows 10. Further efforts are needed to rebuild GSV2SVF for running on Linux.

3. Google Maps API license

A Google Maps API is needed to explore Google Maps and perform GSV queries. The user may apply for a Google Maps API Key at https://developers.google.com/maps/documentation/javascript/get-api-key.

To make sure your Google API is working, you can verify by testing https://maps.googleapis.com/maps/api/streetview/metadata?location=40.7254686,-73.9966184&key=API_KEY Remember to replace "API_KEY" with your actual Google API Key. If it works, you should be able to see response in the following structure:

{
   "copyright" : "© Google",
   "date" : "2019-06",
   "location" : {
      "lat" : 40.7254686,
      "lng" : -73.9966184
   },
   "pano_id" : "zI-rBAALNiHzRTwczq7XDg",
   "status" : "OK"
}

4. Configuration

Open the configuration table Config.csv in the root. The table consists of two columns: the variable names (APIKey, Lat, Lon, CUDA) are laid out in the first column with their respective values stored in the second column. Enter your Google Maps API Key and optionally the startup map location in the second column.

5. Run GSV2SVF (Run.bat)

5.1. Add sample points

The user can add GSV sample points for view factor calculation in three different ways: (1) from the text box; (2) from mouse click; (3) from a shapefile (.shp).

  • (1) Add a single GSV sample point from the coordinate text box. You can either manually type in coordinates in the format of “lat,lon” or by clicking on the map to update the coordinates in the text box, and then click to add the point.
  • (2) Add GSV sample points by interactively double-clicking on the map. First toggle on to enter the interactive sampling mode. To densify the points between two mouse double-click positions, you may toggle on and enter the desired sampling distance (in units of meters).
  • (3) Add GSV sample points from a shapefile. Click to browse for shapefiles and add GSV sample points from the selected shapefile.

5.2. Compute and explore view factors

  • (1) Compute and save the results to the default folder (./Cache/Default) by unchecking and then clicking .

  • (2) Compute and save the results to a separate folder by checking and then clicking . You will be promoted to specify the folder path (relative to ./Cache) to which the results will be saved. When a computation task is done, the sample points will be shaded in colors of red (SVF=0) to green (SVF=1) by their sky view factor values. To view the tree and building view factors at a sample point, you can hover the mouse over the sample point and an information window will pop up displaying the sky, tree, and building view factors together with the unclassified and classified hemispherical street view. The result set for each sample point is stored in a separate folder that contains the following items:

  • (1) The GSV panorama (mosaic.png).

  • (2) The fisheye image (fisheye.png).

  • (3) The classified GSV panorama (mosaic_segnet.png).

  • (4) The classified fisheye image (fisheye_segnet.png).

  • (5) The sample point attributes (sequential number, GSV panorama ID, capture date, latitude, longitude, sky view factor, tree view factor, building view factor). The attributes for each sample point are also written sequentially into a shapefile and a CSV file in the same folder. The shapefile and the CSV file can then be imported into GIS and other tools for further visualization and analysis.

5.3. Export results

Click to interactively define the export extent and then click to export all results within the extent to the specified folder. The exported results are available in two formats, Esri Shapefile (.shp) and comma-separated text (.csv). Both the Esri Shapefile and comma-separated text export formats have the same table schema with eight columns (Figure 8): (1) “id” is the sequential order; (2) “panoid” is the GSV ID returned from the Google Maps API query; (3) “data” is the date of acquisition of the GSV panorama; (4) “lat” is the WGS84 latitude of the location; (5) “lon” is the WGS84 longitude of the location; (6) “svf” is the sky view factor; (7) “tvf” is the tree view factor; (8) “bvf” is the building view factor.

6. Contact

Jianming Liang [email protected]

About

Reduced Version of GSV2SVF to be incorporated into wrapper program

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 86.7%
  • Tcl 4.5%
  • C++ 3.6%
  • QML 2.8%
  • C 1.5%
  • HTML 0.3%
  • Other 0.6%