A Prototype Imaging System with a Biologically Inspired Imager for an Extended Range Walking Stick

This thesis describes the first attempt at simulating a prototype imaging system involving a biologically inspired imager for the purpose of estimating depth information in ambient scenes. This is part of the series of tests intended for the neuromorphic device developed at the ISS Complex Systems Unit under the HAPTIC project, a feasibility study on the integration of neuromorphic imagers with tactile transducers for the real-time transduction of dynamic visual stimuli. It has long been recognized that the focus information contained in an image is an important depth cue for both human and machine vision. This work implements distance estimation by optimizing the amount of image focus on a blurred image in the space of the imaging system parameters where the blurring is due to the imaging system’s point spread function. The defocused image is first restored via blind deconvolution and then the amount of focus from both the original and restored images are calculated using a suitable focus measure. Using a simple golden section line search, the object distance is calculated from the optimal parameters that minimize the square error between the amount of focus from the original image and the reblurred restored image obtained at each optimization step. Results from simulated defocused images are acceptably good in the sense that the algorithm was able to provide good best distance estimates within a narrow range of object distances. However the uncertainties associated to those estimates are quite large, mainly due to blind deconvolution not being able to provide unique ideal image restorations.