Student: Shreyasvi Natraj
The project revolves around finding the possibility to use elevation data related to a particular city before and after undergoing a specific type of disaster in order to be able to do a damage analysis for the area in an automated way. Elevation data can be extracted from DEMs or Digital Elevation Models which are basically latitude, longitude and altitude data extracted from two satellite image stereo pairs for a particular city. This data in combination with building data extracted from shape file can be used to find out the number of damaged buildings in a particular city. A preliminary analysis can also be done based on openstreetmap building data in which depending upon the number of mapped buildings by openstreetmaps, the number of damaged buildings can be found.
This version consists of a openstreetmap scraper (in case building data is to be extracted from openstreetmaps models) and version which uses .csv files extracted from .shp file in order to do damage analysis for all the mapped buildings by comparing their elevation data before and after the event.
Requirements
pip install numpy pandas csv time osmium
The usage of the code can be done in several ways:
Building Data
- Usage with Openstreetmap:
- The building data for a particular city is first downloaded from openstreetmap.org in .osm format and kept in the same folder as the osmreader python programs.
- Run the command
python osmread.py <name of .osm file>to extract all the building coordinate data as well as centroid for each of the buildings.
- Usage with custom .shp file:
- Export the .shp file in .csv format using QGIS or any other external program
Elevation Data
- Elevation data is to be extracted from DEMs which are to be prepared from two image stereo pairs and exporting the raster in xyz file format.
- Elevation data (in xyz file format) can be automatically converted to the required .csv format by running
python pre_proc.pyprogram over the two DEM xyz files (Right now for a 16 million coordinates file, it takes 7.5 minutes to process)
Input file format follows the same format:
| lat | long | elav |
|---|---|---|
| coordinate 1 | coordinate 1 | elav1 |
| coordinate 2 | coordinate 2 | elav2 |
In case of building data the third column would be replaced with metadata NOTE: Coordinates of latitude and longitude for all the files have to be rounded up to 5th decimal point for consistent data
- Run the command
python f_cmp_asc.pyorpython fast_compare.pyin order to run the algorithm over the dataset - When the program is running, you will see cycle numbers, *cycle number will always equal to the number of buildings present in dataset
- In order to run the program in background use
python f_cmp_asc.py > output.log &orpython fast_compare.py > output.log &. - The status of the program can be found using the
toporoutput.logfile by simply opening it using notepad or any other text editor. - The program must automatically run
filter.pywhich is used to get data in an easy to understand format. - The result file can also be imported to QGIS in order to see the damaged areas.
Results were tested for aleppo DEMs
The elevation change is found in terms of percentage where
- If elevation change percentage = -ve, it is increase in elevation at that coordinate point
- If elevation change percentage = +ve, but less than the threshold value, it is usually change due to error in input data
- If elevation change percentage = +ve, but has significant value such as >10, then there is damage to that particular building
The value for threshold has to be set up be comparing these results with the ground truth i.e. pre analyzed areas which underwent damage
Upon uploading the files to QGIS for visualization, the following images shows the building data extracted for the .shp file (in greenish blue colour) and DEM data (in black/purple colour).
Upon implementation of the algorithm, the number of buildings for which the damage was identified is represented follows (in green color):
On categorizing these buildings based on the percentage elevation change, the following results were obtained where the severity of the disaster can be represented from light to heavy using colors blue-green-yellow-orange-red as show below:
The program requires a filter value to be set so as to remove elevation changes that are caused due to error in DEMs or input data. Upon finding the sweet spot for the filter value, a simple if statement can be used in order to remove all the noisy data and only take significant damage from this program.
Full DEMs can be found at: https://goo.gl/4tKZMX