Ph. D. Dissertation Defense - Jiaying Wu - Tuesday, 14 August 2012 10:00am Carlson Auditorium 76-1125

Chester F. Carlson Center for Imaging Science

Ph.D. Dissertation Defense

JIAYING WU

A Signal Processing Approach For Preprocessing and 3D Analysis of Airborne Small-Footprint Full Waveform LiDAR Data

Dr. Jan van Aardt, Advisor

Tuesday, 14 August 2012, 10:00 am

Carlson Auditorium 76-1125

ABSTRACT

The extraction of structural object metrics from a next generation remote sensing modality, namely waveform light detection and ranging (LiDAR), has garnered increasing interest from the remote sensing research community. However, a number of challenges need to be addressed before structural or 3D vegetation modeling can be accomplished. These include proper processing of complex, often off-nadir waveform signals, extraction of relevant waveform parameters that relate to vegetation structure, and from a quantitative modeling perspective, 3D rendering of a vegetation object from LiDAR waveforms. Three corresponding, broad research objectives therefore were addressed in this dissertation. 

Firstly, the raw incoming LiDAR waveform typically exhibits a stretched, misaligned, and relatively distorted character. A robust signal pre-processing chain for LiDAR waveform calibration, which includes noise reduction, deconvolution, waveform registration, and angular rectification is presented. This pre-processing chain was validated using both simulated waveform data of high fidelity 3D vegetation models, which were derived via the Digital Imaging and Remote Sensing Image Generation (DIRSIG) modeling environment and real small-footprint waveform LiDAR data, collected by the Carnegie Airborne Observatory (CAO) in a savanna region of South Africa. Results showed that the preprocessing approach significantly increased our ability to recover the temporal signal resolution, and resulted in improved waveform-based vegetation biomass estimation.

            Secondly, a model for savanna vegetation biomass was derived using the resultant processed waveform data and by decoding the waveform in terms of feature metrics for woody and herbaceous biomass estimation. The results confirmed that small-footprint waveform LiDAR data have significant potential in the case of this application.

            Finally, a 3D image clustering-based waveform LiDAR inversion model was developed for 1^st order (principal branch level) 3D tree reconstruction in both leaf-off and leaf-on conditions. These outputs not only contribute to the visualization of complex tree structures, but also benefit efforts related to the quantification of vegetation structure for natural resource applications from waveform LiDAR data.