Research And Implementation Of Velocity Estimation For UAVS

Quadrotors are now gradually moving toward miniaturization and intelligence.They are often faced with operations in complicated environments.Due to the weak signal of GPS(Global Positioning System),sparse visual texture or insufficient light under such conditions,the conventional state estimation algorithms usually fail which results in the difficulty of stable and safe autonomous navigation.This thesis theoretically analyses the aerodynamic characteristics of quadrotors and reveals that the translation velocity in the motor plane can be observed by accelerometer’s measurements directly.Based on this aerodynamic characteristics,this paper proposed two novel methods to solve the state estimation problem.One is using machine learning method(SVM regression)to train the correlation and predict the velocity from only IMU measurements;the other is a novel state estimator by fusing the aerodynamic model with measurements from optical flow and other low-cost sensors such as IMU,magnetometer and ultrasonic sensor.To make the estimator adaptive to the noise caused by the changing spinning speed of rotors and air density in flight,we also include the drag coefficients in the filter.We also propose a method that can adjust the covariance of optical flow measurements according to light condition around.We have tested out estimator with different platforms in different scenes.Experimental results show that our estimator performs more robustly and better than existing methods in low light or few texture conditions.