Answers are grounded in this paper and cite their sources.
The pose (3D position and orientation) of an Unmanned Aerial Vehicle (UAV) during the landing process aboard a ship can be estimated using a 3D model-based vision system approach. The developed vision system is based on a standard Red, Green and Blue (RGB) camera using one workstation for processing data located on the ship's deck. A ground-based vision system allows the use of small size and weight UAV, due to the low computer requirements onboard. The proposed architecture is based on a Particle Filter (PF) scheme and has three stages: importance sampling, importance weighting and resampling. In the importance sampling, we detect bounding box candidates and apply an appearance-based pose sampler to retrieve the most likely poses. After this, we fuse information from the current frame with information from the previous time step using an Unscented Kalman Filter (UKF) for the translational motion filtering and an Unscented Bingham Filter (UBiF) for the rotational motion filtering. In the importance weighting stage, we use a color based likelihood metric to deal with the expected real sky background filled with clouds. In the resampling stage, we eliminate particles with low importance weights and replicate high weight ones. Results show that performance is compatible with the automatic landing system requirements.
No takes yet. Share an insight, caveat, or question.
Santos et al. (2019) studied this question.