Vision-aided Inertial Navigation Research Articles

Vision-aided inertial navigation system for robotic mobile mapping

A mapping system by vision-aided inertial navigation was developed for areas where GNSS signals are unreachable. In this framework, a methodology on the integration of vision and inertial sensors is presented, analysed and tested. The system employs the method of “SLAM: Simultaneous Localisation And Mapping” where the only external input available to the system at the beginning of the mapping mission is a number of features with known coordinates. SLAM is a term used in the robotics community to describe the problem of mapping the environment and at the same time using this map to determine the location of the mapping device. Differing from the robotics approach, the presented development stems from the frameworks of photogrammetry and kinematic geodesy that are merged in two filters that run in parallel: the Least-Squares Adjustment (LSA) for features coordinates determination and the Kalman filter (KF) for navigation correction. To test this approach, a mapping system-prototype comprising two CCD cameras and one Inertial Measurement Unit (IMU) is introduced. Conceptually, the outputs of the LSA photogrammetric resection are used as the external measurements for the KF that corrects the inertial navigation. The filtered position and orientation are subsequently employed in the photogrammetric intersection to map the surrounding features that are used as control points for the resection in the next epoch. We confirm empirically the dependency of navigation performance on the quality of the images and the number of tracked features, as well as on the geometry of the stereo-pair. Due to its autonomous nature, the SLAM's performance is further affected by the quality of IMU initialisation and the a-priory assumptions on error distribution. Using the example of the presented system we show that centimetre accuracy can be achieved in both navigation and mapping when the image geometry is optimal.

Towards Autonomous Fixed-Wing Unmanned Aerial Vehicle Landing: A Vision-Aided Inertial Navigation under Sensor Reconfiguration Scenario

While autonomous landing of unmanned aerial vehicles (UAVs) requires accurate position estimation, the standard inertial navigation unit (INU, the inertial measurement unit with a global positioning system (GPS)) provides relatively poor accuracy in altitude estimation. A common solution for this problem is to aid the INU with additional sensors and/or ground infrastructures, but the main hurdles to the approach are the limited payload of UAVs and extra cost involved. Dynamic sensor reconfiguration can be a good alternative by constructing a new sensor system utilizing available sensors around without adding new sensory equipment to UAVs. In this paper, a sensor reconfiguration scenario for autonomous fixed-wing UAV landing is considered and the resulting vision-aided inertial navigation system is investigated. This paper presents (i) a sensor fusion algorithm for a passive monocular camera and an INU based on the Extended Kalman Filter (EKF), and (ii) an object-detection vision algorithm using optical flow. The EKF is chosen to take care of the nonlinearities in the vision system, and the optical flow is used to robustly detect the UAV from noisy background. Pilot-controlled landing experiments on a NASA UAV platform and the filter simulations were performed to validate the feasibility of the proposed approach. Promising results were obtained showing 50%-80% error reduction in altitude estimation.

Vision-aided inertial navigation for pinpoint planetary landing

Autonomous and safe landing spacecraft on moon and planetary bodies is a rather difficult and risky task. Accurate relative navigation between the spacecraft and the planetary surface is essential, together with the autonomous hazard detection and avoidance. This paper describes the vision-aided inertial navigation (VAIN) scheme to meet the pinpoint landing requirement of the next generation planetary lander. Images of distinctive surface feature called feature points/landmarks are detected and tracked autonomously to improve the performance of inertial navigation. Landmark image information derived from optical navigation camera and the spacecraft state information sensed by IMU (Inertial Measurement Unit) are integrated in extended Kalman filter algorithm. The validity of the proposed navigation scheme is confirmed by computer simulation.

Vision-Aided Inertial Navigation for Flight Control

Published in the Journal of Aerospace Computing, Information, and Communication, Vol. 2, September 2005