Using finite moment rate of innovation for LIDAR waveform complexity estimation

LIDAR sensing collects data to obtain detailed topographical information about a region. A major challenge here is the large amount of data that needs to be collected for accurate surface reconstruction which in turn imposes significant storage, processing and transmission requirements. Current efforts to solve this problem have been focusing on applying compression algorithms to the range measurements. This process, however, requires the collection of large amounts of data, most of which is ultimately discarded. Instead, our approach is to apply a compressive sensing paradigm to sparsely sample the scene using the estimated pulse complexity to characterize the scene complexity, thus determining the number of samples needed for accurate reconstruction. As a first step towards this goal, we characterize here the individual laser return pulses for scenes of varying complexity, assuming that the pulses are composed by a finite number of innovations. Each scene is thus sampled at a rate that is much lower than the traditional Nyquist/Shannon limit. Our results show that accurate classifications of waveform and scene complexity with the simple nearest neighbor classifier are obtained using the proposed algorithm.