building-feature-recognition

In this repository, we accomplished building feature recognition using traditional/dl-assisted computer vision method. The Chinese version of the README is here. And the report of our project(in Chinese) is in the report folder.

Results

number of floors

number of windows

size of windows

area of windows($m^2$)

max length($m$)

max width($m$)

max height($m$)

Floor area($m^2$)

volume($m^3$)

Using the code

To use our code, please first clone this repository and install the cv2, numpy, matplotlib package.

Reproduction

side_main.py process pictures taken horizontally and count floor numbers as well as their relative proportion.

top_main.py process pictures taken by a drone from the top of the building.

window_main.py process pictures taken horizontally and count window numbers as well as their relative proportion.

Simply click the 'run' icon when opening the three files and you will see the results. Note that threshold tuning might be needed, please refer to our report for more details.

The demo pictures and the results are in the demo folder.

Test your own picture

Please note that the computer vision methods are sensitive to the quality of the picture.

To test your own picture, you may simply replace the picture name in the code. If the default thresholds don't work well, just replace it with a few tests.