Design Of A Beidou/GPS Quadrotor With Vision-Based Landing

Quadrotor is becoming a research focus in the field of navigation and control recently since its simple structure,small size and high maneuverability.Generally,most navigation system in a quadrotor is organized with Global Position System(GPS),however,considering the system security and proprietary intellectual property rights,it is an urgent requirement that the Beidou navigation system should be involved in quadrotor navigation system.On the other hand,vision-based landing technique in quadrotor not only improves the level of autonomous flight,but broadens the scope of application.Therefore,it is of great significance to study the vision-based landing algorithm.Firstly,this thesis proposes a quaternion-based Kalman filter for AHRS with Adaptive-Step Gradient Descent(ASGD)algorithm.In compare with the standard EKF,ASGD is more computationally simple and be free of interference stimulated by magnetic distortion.A inertial/Beidou/GPS fusion algorithm is also designed to implement inertial/GNSS integrated navigation technology.Secondly,to decrease the calculating amount,a reduced computational single-corner filter is described to replace the local non-maximum suppression in Harris corner detection algorithm.For the same reason,in the processing of object recognition,a fast and reliable square recognition algorithm is also presented.Finally,a carbon fiber quadrotor,an anti-air disturbance and box-like build-in damping flight controller,a vision processing module and a vision debugging module are constructed.A Handheld Ground Station(HGS)is also established to debug parameter online.Comparing AHRS algorithm proposed in this thesis with the standard EKF based on these experimental hardware platforms,it is seen that the AHRS runs faster than EKF and be free of magnetic distortion.The real-time object recognition experiment shows that the square recognition algorithm presented in this thesis is able to identify the target image accurately at 3Hz update rate.The vision-based landing in the real world was also implemented in the end.