Wavelet Transform Based 3D Scattered Data Processing in Binocular Micro Stereovision System

Abstract

In micro stereovision system, false matching due to the randomicity of image signal makes 3D reconstruction data contain a large amount of abnormal data in the form of noise (pulse noise, Gauss noise). In order to obtain a more accurate 3D reconstruction shape, a novel method of 3D scattered data processing, which is a combination of wavelet transform and extended Vector Median Filters, is proposed to use wavelet transform to realize primary denosing, and use extended Vector Median Filters to further denoise thereby achieving higher precision 3D measurement. In wavelet based denosing, wavelet based edge detector in vector space removes edge abnormal points effectively, moreover, boundary tracking is used to prune out the image background from the matching area to remove exterior abnormal points. Then the local areas determined by edges are filtered by extended vector median filter. Experimental results indicate that the method proposed is effective to removing abnormal data from the 3D reconstructed data and enables 3D measurement position accuracy to reach 3 m µ .