A model-based approach to computer vision and automatic control using Matlab Simulink for an autonomous indoor multirotor UAV

Abstract

A prototype autonomous UAV platform featuring computer vision based navigation for use in GPS denied environments is presented.The UAV is based on a hexacopter platform from 3DR Robotics which has been equipped with a PandaBoard ES single-board computer and a downward facing webcam.Position estimation is performed using template and feature point matching image analysis techniques and object recognition has been implemented using invariant moment descriptors.Position control has been achieved using cascaded PD control, generating attitude setpoints sent to the hexacopter on-board computer over Ethernet.Image analysis, control and decision making algorithms have been developed using modelbased design techniques with automatic code generation in Matlab Simulink.A multirotor model has been obtained with system identification methods and a camera model and emulator have been developed and used to emulate a camera video feed for image analysis algorithms development and verification.A 3D-visualization environment has been developed and used for assessment of the simulated system performance and behavior.Model accuracy is considered high, image analysis and control algorithm parameters tuned in simulation give similar flight behavior during actual test flights.The UAV prototype is capable of limited time hovering above a play mat floor surface.Insufficient hexacopter altitude and yaw control performance achieved by the hexacopter computer does, however, affect position estimation in a negative way and arguably making it perform unsatisfactory.The template matching position estimation technique is functional but image feature point matching methods should be considered in future development for improved position estimation robustness to hexacopter yaw and altitude change.