A geometric model for fusing IMU into monocular visual localization of 3-D mobile robots

For the localization problem of mobile robots in 3D space, vision-based methods are being extensively investigated in recent years, due to the information richness provided by images. To improve the robustness and continuity, usually more sensors are introduced to assist visual localization in practices. In this paper, we propose a geometric model of fusing IMU data into the visual tracking process to estimate the robot trajectory in 3-D space. The model employs tracking of a certain number of feature points in the environment, and fuses them with sensor data from IMU to perform iterative adaptive estimation of both the robot trajectory and the positions of the selected feature points. The convergence and performance of the proposed method is validated in theoretical proof and simulation test.