Micro Quadrotor Navigation Based On Image Processing

Recently rotor unmanned aerial vehicles has been applied in many areas.Whether to perform reconnaissance combat mission in the military field,or to do the search and rescue work of disasters in civil field,navigation of rotor unmanned aerial vehicles plays a crucial role.The inertial navigation equipment and GPS(Global Positioning System)or other satellite navigation equipment are widely applied in traditional navigation methods,and GPS navigation are the most widely used equipment.But for many places that has no GPS signals or are not suitable for GPS application,other navigation methods need to be urgently developed.Computer vision technology for image processing can get rich information,so its application in the rotor-unmanned navigation has great potential.Aiming at the problem of rotor unmanned aerial vehicle(UAV)in visual navigation,image matching technology is used to solve the actual position of UAV.In this thesis,the method of information fusion between reference map and motion estimation is proposed to locate and navigate.The image feature points extraction and image matching are analyzed.The image feature points are extracted by ORB algorithm,and the descriptor which are taken as the reference map with the corresponding geographic coordinates is obtained.At first each reference map is collected with the size of 640 * 480.Then the descriptor and the corresponding coordinates in the form of text files are saved to the computing unit.The index table that contains picture number,picture GPS and picture angle is made and then stored to the local.The index table is sorted by longitude and latitude for matching and searching.For the matching problem of feature points,the fast LSH algorithm is adopted,and the error of result is processed by RANSANC.In order to improve the search matching speed and increase the practicability of visual navigation that is mentioned in this thesis,the KLT feature point tracking algorithm is used to estimate the motion information of the rotor UAV,and a rough position for searching reference map is obtained.In this thesis,we calculate the relative offset between the current and previous frames,convert the offset from the picture coordinate system to the world coordinate system,and use the offset accumulated from the starting position as the current position of the UAV.The picture within 15 meters of the rough position is selected as candidate match graph,which will narrow the search range and improve the real-time of navigation.And the exact position obtained by matching reference map is used to update the initial position of the estimated motion information algorithm,thereby reducing the accumulated error.Aiming at the practicality of the algorithm,this thesis designs a complete visual navigation software system and completes the test on the Dajiang M100 Quad-rotors unmanned aerial vehicle.The ROS robot operating system is the basic framework of the software,and the other parts are prepared in a modular way,including the real-time map to extract the feature points and the reference map to search and match and the four-rotor UAV position estimation.The information exchange between the modules is completed by ROS message communication mechanism.The four-rotor UAV flew in a 200-meter air flight mission,which is to fly according to the preset waypoint.