State Estimator using Hybrid Kalman and Particle Filter for Indoor UAV Navigation

Abstract

Unmanned aerial vehicles (UAVs) are being used for outdoors inspection and surveying tasks. When operating in an outdoor environment, the global navigation satellite system (GNSS) is predominantly used for position aiding, and magnetometers are used for heading aiding. In combination with an inertial sensor, these sensors form the backbone for state estimation for drones operating in an outdoor environment. A desire to utilize UAVs for inspections in indoor environments means that new challenges are faced. One of these challenges is that the traditional GNSS is unavailable for position aiding, and magnetometers can be unreliable in the presence of industrial equipment. This thesis aims at proposing, developing, and implementing a filtering solution capable of performing indoor autonomous navigation. A Hybrid filter solution is proposed where the GNSS and magnetometer are replaced by a stereo camera for depth perception. The Hybrid-filter is composed of a Kalman filter loosely coupled with a Particle filter. The Kalman filter is the main navigation filter in this framework. The navigation solution is based on integrated inertial measurements and aided by position and heading estimates from the Particle filter. In turn, the particle filter utilizes the velocity and attitude estimates from the Kalman filter, along with the depth data from the stereo camera to estimate the position and heading of the UAV.