We present a novel real time 3D Automatic Target Recognition algorithm appropriate for LIDAR based time critical applications. Its main contribution is the Constant False Alarm Rate adaptive threshold combined with the Projection Density Energy and the transformation of the 3D problem into multiple 2Ds. Our approach is invariant to 3D rotations combined with scale change, Gaussian noise and uniform sparse representation of the target.
Applied on real targets from the UWA dataset and on military targets from the Princeton shape benchmark, we obtained 90% recognition in 77ms and 97% in 106ms respectively (in Matlab). Our approach could be considered by the Defence Community as an initial step towards LIDAR based missile seekers where data is inversely proportional to the available time to perform recognition.