A sensor fusion method to solve the scale ambiguity of single image by combining IMU

Abstract

Simultaneous Localization and Mapping, also known as SLAM, is a challenging problem in the mobile robotics community. It deals with building a map of an unknown environment and tracking the pose of a robot relative to the physical environment at the same time with a sequence of measurements obtained from various sensors. This paper proposes a method for fusing IMU with a single camera to overcome the scale uncertainty problem of the SFM-based SLAM. The key notion of the proposed approach is to use distance (translational) information for a robot estimated from an IMU sensor to correct (or compensate for) the translational information, which has universal scale ambiguity from the SFM and to complete the projection matrices that will be used to estimate the scale. To enhance the accuracy of the scale estimation, the back-projection and verification method is used. An experimental study was done to evaluate the viability of the proposed technique and to serve as an extendable framework for SLAM.