Image classification: recognizing various features of many buildings from images

Question

So, let's say I have the longitude/latitude or address of many buildings

I can get satellite images, "street view", and perhaps 3d/perspective views of buildings.

I want to find the:

height, number of floors, floor area (max building footprint)

of the building. about 200k buildings.

Is there a library for recognizing buildings from satellite shots or pictures? Kind of like face detection I suppose.

Any other suggestions?

Thanks!

Answer 1

Yes and no. You will have to program many of the feature recognition yourself, but this should just be edge detection and point-of-interest detection (corners, straight lines, grids) to work out how many windows high a building is.

To do this you might want to checkout OpenCV (Computer Vision) which has C routines for real-time image analysis and common operations such as edge detection. I can also recommend DIP3e as a book to read and Image Processing in C and it's Java counterpart (if Java floats your boat).

Image processing is fun, and this sounds like a great project. There is tons on the Internet to learn from ... check out AI techniques to train image recognition systems to get better using a known sample set. Fun times!

Also, be careful of perspective ;)

good luck