Aural imaging from 3D vision

Abstract

Dense stereo maps have many useful applications in computer vision. They can help in performing such tasks as background separation, segmentation, and object recognition. Despite the continued exponential advances in the computational capacity of microprocessors, current personal computers are limited in their ability to adequately solve the dense stereo vision problem in realtime. Modern processors now contain a SIMD instruction set architecture that can provide substantial improvements in performance when working with specific data types. This program demonstrates a stereo vision algorithm that implements commonly known methods in computer vision and takes advantage of the Altivec instruction set to achieve realtime performance. Once the stereo data has been processed, the depth map that is produced can be used to provide an additional dimension of data to the user. The software demonstrates a possible use for this data, as an aural aid to people that are blind or vision impaired. The system uses the depth information to identify obstacles in a user's path and alert the user with aural feedback.