Automatic Recognition of Buildings in Satellite Imagery This project automatically detects buildings in satellite images. It uses the Burns Edge Detection algorithm to detect edges of buildings.
Algorithm uses Burns algorithm to improve edge detection in a rasterized image. Example of initial corner and edge detection is shown below.
Algorithm eliminates detects all small line segments, groups similar line segments into larger lines. Intsersections of lines at specified angles consistute potential building corners.
Algorithm reviews the lines and corners and eliminates redundant edges
A building shape is found. Further work needed to improve the algorihm's detection rate.
Image is pre-processed with Gaussian smoothing to remove some image noise.
Burns edge detection is run to look for gradients in image that indicate edges. This algorithm sorts the lines into “gradient bins” based on the angle of the line. The algorithm then identifies the most dominant lines based on size
Line regions are constructed into lines based on a least squares fit of a line of all points in that region
Lines are then compared and linked if they are sufficiently close in lateral distance, angle, overlap, and underlap
Resulting line segments are then compared to extract all possible line intersections. Note that a tolerance is required here since some line segments do not completely overlap at the building corner point.
From these line intersections, building corners are determined as those which intersect at an angle close to 90 degrees. Note this 90 degrees tolerance can change drastically based on the acquisition angle of the imaging satellite. The result of this stage is a series of x,y points where a building corner has been identified, as well as information on the two walls that intersect it.
Algorithm looks at all possible building corners and identifies sets which close a loop.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent