Harsh-Environment Visual Odometry for Field Robots Using Data Fusion of Gyroscope & Magnetometer

Abstract

This paper presents a harsh-environment visual odometry method that is robust to the robot orientation as well as the camera movement in an outdoor environment. The accuracy of visual odometry in robots can be enhanced by using additional sensor measurements such as an encoder, gyroscope, and/or magnetometer. This strategy can even reduce the computational time. However, in an outdoor environment, the moving robot can experience vibration, which causes unsynchronized data fusion between the camera and additional sensors. This unsynchronized data fusion causes errors in the robot orientation, which can lead to unwanted large drift errors in localization. To overcome this problem, firstly two distinctively different characteristics of the gyroscope and the magnetometer are combined to estimate the robot orientation. The initial robot orientation is estimated by integration of the gyroscope input, and this initial robot orientation is corrected using the magnetometer data in bundle adjustment. Secondly, the poses of the robot and the camera are estimated separately, and these separately estimated poses of the robot and the camera are used in feature matching and bundle adjustment to reduce drift errors in localization in the outdoor environment. The performance of the proposed method is demonstrated using dataset-based experiments. Copyright (C) 2020 The Authors.