Self-localization of Mobile Robot Based on Binocular Camera and Unscented Kalman Filter

Abstract

Self-localization is a fundamental requirement for a mobile robot. In indoor environments, the objects are polygonal usually. These objects can be described as line segments. Depth image of the environment can be obtained by a binocular camera automatically. Clustering technology and least-square method have been used to extract features of line segments. The system model and the observation model have been established. Extended Kalman filter (EKF) is the standard method for parameter estimation and information integration. But the EKF has its flaws. A nonlinear system is linearized to a linear system. So the accuracy of the EKF can only reach to first-order. And the EKF needs to calculate Jacobian matrices. In order to overcome the disadvantages of the EKF, unscented Kalman filter (UKF) has been used to integrate the data from the odometry and the binocular camera to obtain the accurate pose of the mobile robot. It is proved by experiments that the algorithm based on the UKF is obviously more accurate than the algorithm based on the EKF.