Pose estimation of a mobile robot using monocular vision and inertial sensors data

Abstract

Practically, today most mobile devices, such as robots require to have ability to determine their pose (location and orientation) from low-cost sensors with high accuracy. This paper presents a novel fusion method to determine the pose estimation of an autonomous mobile robot using inertial sensors and a camera. Speeded up robust features (SURF) is used as a visual algorithm to detect natural landmarks (also known as markerless method) from image sequence. SURF is an effective detector and descriptor algorithm which can detect key point features from images regardless of the lighting or different viewing conditions. The inertial sensors used are six degree of freedom (6DOF) which comprises 3-axis of accelerometer and 3-axis of gyroscope. The data from inertial sensor and vision are fused together using Extended Kalman Filter (EKF). The key contribution of this paper is the low-cost, easy deployment and fast computation. The system combines the best of each sensor, more information derived from the camera and the fast response of the inertial sensors. Experimental and simulated results show that this method is fast in computation, reliable and improves accuracy. Root mean square errors (RMSEs) for position and orientation were achieved in the experiments.