3D classification of through-the-wall radar images using statistical object models

For a variety of reasons it is desirable to know who or what is located behind a wall from a stand off distance. Radar has been shown to be the most effective sensor for this task. Research so far has primarily focused on image formation. However, imagery obtained from raw back scatter data is not easily interpreted. In particular, clutter masks the presence of real objects. At a minimum, we need a tool that would produce an occupancy map of the behind-the-wall scene using machine-assisted interpretation. This work reports on the development of such a tool. Object classes are developed from collected 3D data. The Mahalanobis distance is the metric that is minimized to assign samples from the volume image to their respective class. Labels along three spatial dimensions are thenfused to produce afinal labelfor 3D cells. Thefinal result is the interpreted volume image of behind-the-wall scene occupancy.