Street Segmentation from QGis rasters

How to interact

The images from the diferent sensores should be in a forlder named R10m. The images can be obtained here and the files are located at GRANULE/L2A_T29TNF_A005497_20180326T112919/IMG_DATA/R10m;
Simply run Street_seg.py and two windows will be displayed.
The left one shows the original satellite raster preview
The right one displays the results obtained after segmentation.

Some visual results are presented at Resultados folder.
Only the TCI raster is used since it is an RGB image built from the B02 (Blue), B03 (Green), and B04 (Red) Bands. Therefore there's no need of reading the images of the 3 individual sensors.
Because the input is based on a truly HDR image, every time the zoom (scale) of the windows are changed, the displayed portion of the raster is normalized for an RGB scale between [0,255]. This is done by taking into account the maximum value of the pixels of the displayed portion, allowing a dynamic adjustment based on the light intensity of that portion.
For higher values of zoom, the original raster presents already some blur. In these scenarios, applying a gaussian blur filter to reduced soft edges before canny is irrelevant and not useful to the final result.
An adaptive cauny method is used to detect the edges of the raster. The lower and upper limits are estimated based on the median of the image as seen here.
After applying the canny filters, the green (chlorophyll) zones are "cleaned" from the result. This is made base on an analysis of the pixel colour in the HSV space and allows to remove edges encounter in places like rivers surrounded by green zones or edges between field crops.

Tecnologias e Aplicação de CG - Cartografia

Mestrado Integrado em Engenharia Informática

Departamento de Informática

Universidade do Minho - 2018