Robust L ∞ convex optimisation for UAVs cooperative motion estimation

In this paper, a system for real-time cooperative monocular visual motion estimation with multiple unmanned aerial vehicles is proposed. Distributing the system across a network of vehicles allows for efficient processing in terms of both computational time and estimation accuracy. The resulting global cooperative motion estimation employs state-of-the-art approaches for optimisation, individual motion estimation and registration. Three-view geometry algorithms are developed within a convex optimisation framework on-board the monocular vision systems of each vehicle. In the presented novel distributed cooperative strategy a visual loop-closure module is deployed to detect any simultaneously overlapping fields of view of two or more of the vehicles. A positive feedback from the latter module triggers the collaborative motion estimation algorithm between any vehicles involved in this loop-closure. This scenario creates a flexible stereo set-up which jointly optimises the motion estimates of all vehicles in the cooperative scheme. Prior to that, vehicle-to-vehicle relative pose estimates are recovered with a novel robust registration solution in a global optimisation framework. Furthermore, as a complementary solution, a robust non-linear H_∞filter is designed to fuse measurements from the vehicles’ on-board inertial sensors with the visual estimates. The proposed cooperative navigation solution has been validated on real-world data, using two unmanned aerial vehicles equipped with monocular vision systems.