OKVIS

OKVIS stands for Open Keyframe-based Visual-Inertial SLAM. It is an open-source framework for visual-inertial odometry and SLAM designed for use in robotics, drones, and mobile devices. The system aims to provide robust motion estimation by fusing camera imagery with inertial measurements.

The core idea of OKVIS is to combine data from monocular or stereo cameras with an inertial measurement unit (IMU) to estimate the device’s trajectory and build a map of the environment. It relies on a sliding-window nonlinear optimization that jointly estimates camera poses, IMU biases, and three-dimensional landmark positions. IMU preintegration is used to efficiently incorporate high-rate inertial data between keyframes, enabling accurate fusion with visual observations. The approach treats the problem as a state estimation task that balances information from image correspondences and inertial measurements while applying robust estimation techniques to mitigate outliers.

In operation, a front end handles feature tracking and frame-to-frame correspondence, while a back end maintains a local map and performs the optimization over a window of recent frames. This keyframe-based strategy helps manage computational complexity and supports real-time performance on typical embedded or workstation hardware. OKVIS provides data products such as pose estimates, trajectory, and a sparse map of landmarks, which can be used for localization, navigation, or further mapping tasks.

OKVIS is distributed as open-source C++ software with interfaces for common robotics middleware such as ROS. It has been adopted in academic research and educational settings as a reference implementation for visual-inertial fusion and SLAM, and it has influenced subsequent visual-inertial SLAM systems.