Subpixel disparity for accurate stereoscopic depth and velocity

Abstract

This paper presents a subpixel disparity technique for obtaining accurate depth & velocity estimation of a moving object using stereoscopic vision. In such system, depth information is obtained from calculating the disparity between two images. This value is often precise within )1 pixel. The limitation in pixel precision leads to inaccuracy in getting 3D information of an object. We propose a technique to increase the precision of disparity calculation by interpolating the values of Sum of Absolute Difference (SAD) within the matching histogram, so as to provide a more accurate depth information, which subsequently gives a more accurate value for estimating velocity of a moving object. The techniques of Least Square Method (LSM) and Doolittle factoring is utilized to perform interpolation, and solve the resulting system of linear equations. Experiments show a significant increase in the accuracy of approximating depth and velocity of a moving object.