Stereo Vision Based Distance Measurement and its Application

Abstract

In computer graphics, Stereo Vision has been a research hotspot for many years, and it is been widely used in many areas. Stereo Vision is a technique which utilizes computer to simulate human eye system. In order to achieve this simulation, two main problems need to be solved: camera calibration and stereo matching. We focus on stereo matching in this paper. After years of development some achievements have been made in stereo matching, but some problems remain unsolved. The two most important things concerned in stereo vision trend to be contradictory: accuracy and efficiency. This paper presents a method called Advanced-Census which is a good combination of SAD and Census. When finding corresponding pixel in left and right image, SAD can get high accuracy with low speed and Census have the opposite result. Advanced-Census has advantages of both SAD and Census. It retains the speed of Census while having the accuracy of SAD. Although it has the speed of Census but not enough for practical applications, so we speed up Advanced-Census using multi-thread technique and edge detection. After speeding up, Advanced-Census gets nearly real-time performance.