Landing System Development Based on Inverse Homography Range Camera Fusion (IHRCF)
Open Access
Abstract
The Unmanned Aerial Vehicle (UAV) is one of the most remarkable inventions of the last 100 years. Much research has been invested in the development of this flying robot. The landing system is one of the more challenging aspects of this system’s development. Artificial Intelligence (AI) has become the preferred technique for landing system development, including reinforcement learning. However, current research is more focused is on system development based on image processing and advanced geometry. A novel calibration based on our previous research had been used to ameliorate the accuracy of the AprilTag pose estimation. With the help of advanced geometry from camera and range sensor data, a process known as Inverse Homography Range Camera Fusion (IHRCF), a pose estimation that outperforms our previous work, is now possible. The range sensor used here is a Time of Flight (ToF) sensor, but the algorithm can be used with any range sensor. First, images are captured by the image acquisition device, a monocular camera. Next, the corners of the landing landmark are detected through AprilTag detection algorithms (ATDA). The pixel correspondence between the image and the range sensor is then calculated via the calibration data. In the succeeding phase, the planar homography between the real-world locations of sensor data and their obtained pixel coordinates is calculated. In the next phase, the pixel coordinates of the AprilTag-detected four corners are transformed by inverse planar homography from pixel coordinates to world coordinates in the camera frame. Finally, knowing the world frame corner points of the AprilTag, rigid body transformation can be used to create the pose data. A CoppeliaSim simulation environment was used to evaluate the IHRCF algorithm, and the test was implemented in real-time Software-in-the-Loop (SIL). The IHRCF algorithm outperformed the AprilTag-only detection approach significantly in both translational and rotational terms. To conclude, the conventional landmark detection algorithm can be ameliorated by incorporating sensor fusion for cameras with lower radial distortion.Keywords
This publication has 44 references indexed in Scilit:
- Control of fixed-wing UAV at levelling phase using artificial intelligenceIOP Conference Series: Materials Science and Engineering, 2018
- Pose Estimation of a Mobile Robot Based on Fusion of IMU Data and Vision Data Using an Extended Kalman FilterSensors, 2017
- Autonomous landing of multi-rotors UAV with monocular gimbaled camera on moving vehicle2017
- Monocular vision-based real-time target recognition and tracking for autonomously landing an UAV in a cluttered shipboard environmentAutonomous Robots, 2016
- Semi-direct EKF-based monocular visual-inertial odometry2015
- 3D Convolutional Neural Networks for landing zone detection from LiDAR2015
- Autonomous Robotics and Deep Learning2014
- Combining Stereo Vision and Inertial Navigation System for a Quad-Rotor UAVJournal of Intelligent & Robotic Systems, 2011
- A bimodal approach for GPS and IMU integration for land vehicle applications2003
- A Method for Constructing Local Monotone Piecewise Cubic InterpolantsSIAM Journal on Scientific and Statistical Computing, 1984