Disclosed is a method for 3D building extraction, comprising: extracting building footprints from one or more stereo images for a building; determining, from rational polynomial coefficient of the stereo images, first height estimation of the extracted building footprints; obtaining, from multi-temporal SAR images for the building, scatters with stable attributes; determining second height estimation for the determined scatters; and combining the first height estimation and the second height estimation to generate a fused height for each of the extracted building footprints.