Algorithms for 3D time-of-flight imaging

Abstract

This thesis describes the design and implementation of two novel frameworks and processing schemes for 3D imaging based on time-of- flight (TOF) principles. The first is a low power, low hardware complexity technique based on parametric signal processing for orienting and localizing simple planar scenes. The second is an improved method for simultaneously performing phase unwrapping and denoising for sinusoidal amplitude modulated continuous wave ToF cameras using multiple frequencies. The first application uses several unfocused photodetectors with high time resolution to estimate information about features in the scene. Because the time profiles of the responses for each sensor are parametric in nature, the recovery algorithm uses finite rate of innovation (FRI) methods to estimate signal parameters. The signal parameters are then used to recover the scene features. The second application uses a generalized approximate message passing (GAMP) framework to incorporate both accurate probabilistic modeling for the measurement process and underlying scene depth map sparsity to accurately extend the unambiguous depth range of the camera. This joint processing results in improved performance over separate unwrapping and denoising steps.