Abstract
The fusion of visual and inertial measurements for motion tracking has become prevalent in the robotic community, due to its complementary sensing characteristics, low cost, and small space requirements. This fusion task is known as the vision-aided inertial navigation system problem. We present a novel hybrid sliding window optimizer to achieve information fusion for a tightly-coupled vision-aided inertial navigation system. It possesses the advantages of both the conditioning-based method and the prior-based method. A novel distributed marginalization method was also designed based on the multi-state constraints method with significant efficiency improvement over the traditional method. The performance of the proposed algorithm was evaluated with the publicly available EuRoC datasets and showed competitive results compared with existing algorithms.
Highlights
Accurate localization in an unknown environment is essential for a robot to succeed in its missions.In many cases, existing external localization systems, such as motion capture systems, the global positioning system, or a pre-constructed map of the working area, are costly, insufficiently accurate, or unavailable
We propose a novel hybrid sliding window optimizer (HSWO) that has the advantages of both the conditioning-based and prior-based methods
We evaluated the proposed hybrid sliding window optimizer on the publicly available EuRoC
Summary
Accurate localization in an unknown environment is essential for a robot to succeed in its missions. After the measurements of IMU are added, the first node (including the pose, velocity, and biases of IMU) of the sliding window must be fixed to eliminate the ambiguity of motion [4] This kind of method is highly robust but is not optimal theoretically, because only a part of the graph is active while LBA is performed. For the prior-based method, a marginalization technique is performed on the edges related with the outside nodes to construct a prior distribution (typically, a Gaussian distribution) for the nodes in the sliding window This kind of method is optimal theoretically but is affected by linearization errors numerically and cannot cope properly with features whose tracking length lies outside the range of the sliding window.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have