| This research project explores full-waveform analysis of three-dimensional "flash" laser radar (ladar) signals for the purpose of segmenting and classifying ladar images. A novel nonparametric range data segmentation designed for experimentally collected ladar data is presented, and yields results superior to conventional edge and boundary detection methods. The completeness of the segmentation enables calculation of additional target characteristics, such as the orientation angle and reflectance of the illuminated objects. Although a cross-correlation ranging algorithm is used to process the collected ladar data, the additional information gained from the segmentation routine is a new contribution to understanding the range of capabilities of the ladar camera used throughout this research.;In addition to the nonparametric segmentation, a widely accepted ladar waveform model is updated to include the effects of optical backscattering from elongated targets. The effective gain of the waveform model accounts for target incidence angle, and shows agreement with independently published research. A successful waveform analysis is achieved by estimating signal amplitude, range, and bias, and the results are compared to other accepted methods.;Additionally, local statistics of the waveform and object parameters are used in an unsupervised nonparametric classification routine. Local statistics have been used frequently in image analysis to differentiate between both objects and regions, and this approach is attempted here. Feature selection of both the segment point data and higher-order statistics emphasizes intrinsic material properties, given the emphasis on classifying similar objects together. Two-sample rank tests of location and dispersion form the core of the classification method. |
