In order for a building detection and/or reconstruction algorithm to make use of features from multiple data sources, those sources must be registered (projected onto one another or describable by a single coordinate system). Because both LiDAR data and a single, overhead aerial image are either typically available or procurable for most desired terrains, these are the data sources considered.

If both aerial image and LiDAR data exist for a given scene, then it behooves one to take advantage of the information existent in both sources of data for more accurate building detection and/or reconstruction. Several building detection/reconstruction approaches existent in the literature treat the registration of the two data sources as a pre-processing technique and proceed to manually select corresponding control points in both sources and then perform the transformation projecting one image onto another. With the plethora of image registration methods published, several of which are automated, it is possible to instead automatically carry out this task.

The proposed registration scheme is implemented in the following steps: (1) detect buildings in LiDAR data; (2) interpolate LiDAR to fixed point spacings; (3) detect buildings in aerial imagery; (4) register detected buildings in LiDAR interpolated range image to aerial image.  The above registration paradigm is implemented in the following block diagram:

The phase only matched filtering (POMF) registration algorithm automatically extracts the geometric transformation parameters from the Fourier transform space. Therefore, in order to use the algorithm, a two-dimensional Discrete Fourier transform of both the aerial image and the LiDAR data must be taken. The irregular LiDAR data must therefore be interpolated to fixed point spacing. The following interpolation scheme is therefore proposed:

The proposed algorithm, after extracting the geometric transformation parameters from the Fourier transforms of the binary images, applies those transformations to only the building areas in the range image thus registering those building areas on top of the aerial image as seen below:

Fore more on this research effort please refer to the following conference paper and presentation: