Stereo Visual Inertial Odometry Using Incremental Smoothing

Visual inertial odometry (VIO) is a technique which estimates the motion of a mobile platform using the measurements from visual and inertial sensors. Recently, these sensors are made available with higher accuracy, smaller size, but lower cost thanks to the developing manufacturing of chips and micro-electromechanical systems (MEMS) devices. However, it is very challenging for real-time visual inertial odometry to obtain robust and accurate localization and mapping with the limit of computational resources. In this paper, we present a stereo visual inertial odometry which pre-integrates IMU measurements to reduce the variables to be optimized and avoid repeated IMU integration during optimization. In addition, incremental smoothing is employed to obtain Maximum A Posteriori (MAP) estimates. The proposed method is shown to achieve high accuracy and efficiency in simulation results.

1 Introduction

Localization and mapping are two major topics in the research area of robot autonomous navigation and positioning, which is essential for many applications, such as autonomous cars, search and rescue, as well as household robots. Visual odometry (VO), an important approach to localization and mapping, is hard to obtain a robust solution as the images captured by the camera are easily affected by environment conditions and the results are sensitive to the noise of measurements. Since inertial measurement units (IMUs) can typically acquire accurate short-term measurements at high rate and are resistant to external interference, combining visual and inertial measurements has been a popular means to compensate for the errors made by IMUs and cameras, and eventually, provide more accurate estimates for localization and mapping.