A Low-Cost Stereo Vision System for Real-Time Pose Estimation and Its Application for Robot Tracking

摘要

In this paper, a 3-D pose estimation system by using stereo vision with low-cost devices is presented. It is developed as a base system for application development. Two webcams and a planar target with circular markers are used to reduce development cost and computational complexity. To avoid correspondence search problem, user has to select regions of interest (ROI's) of each marker on the two images in the same sequence before starting the 3-D reconstruction process. Linear triangulation method is applied for 3-D position calculation of each marker. These positions and the positions of the markers referenced in the planar target coordinate frame are used for pose estimation by using least-squares fitting algorithm to obtain the position and orientation of the planar target. The system can be applied for robot tracking as shown in the experiments. The experimental results validate the system's ability to estimate object pose in real-time with minimum system frequency of 25 Hz.