Research On Key Technology Of Stereo Visual Obstacle Avoidance And Autonomous Navigation For Micro Air Vehicles

In the complex and dangerous environment of low altitude and indoors,the micro air vehicles(MAV)must evade the threat or obstacle of the ground and air,and the limitation of its capacity and carrying weight will bring great difficulties to the research of the related technology of the obstacle avoidance and autonomous navigation.The core of autonomous obstacle avoidance and safe flight of MAV is based on two aspects: one is the real-time acquisition of aircraft's own absolute position,velocity and attitude,as well as the other side is accurate measurement of the relative information of the surrounding environment and potential collision threat objects.In order to achieve the goal of high precision autonomous navigation and positioning capability,the MAV can also realize the three-dimensional modeling of the surrounding environment,and to distinguish and monitor the obstacles that may be encountered on the flight path,and then make a reasonable avoidance strategy to ensure the completion of the established task.In this paper,based on the motion characteristics and low altitude flight environment of MAV,the key technology of three-dimensional environment modeling and obstacle avoidance is carried out,and the research on the method of autonomous navigation based on inertial/stereo visual odometer is carried out.In this paper,a set of binocular stereo vision obstacle avoidance method and integrated navigation method suitable for MAV are presented to provide reliable sensing ability for flight environment and itself motion state information,and provide feasible scheme and reference for realizing the safe flight and autonomous navigation of MAV in complex and dangerous environment.The three-dimensional reconstruction method based on binocular stereo vision is studied in order to realize the flight environment perception of MAV effectively.In view of the problem that the binocular camera has the error matching on edges and the depth information missing in the actual use,based on the traditional idea of the depth image optimization of the combined bilateral filter,a deep image optimization method based on guide filtering is designed,which can be used to optimize the depth image quality.While using the stereo vision and three-dimensional reconstruction technology to enhance the ability of micro-aircraft to perceive the environment,the relative distance information between the obstacles and MAV can be obtained in real time.Based on the calibration results of binocular camera and the binocular ranging model,the error characteristics of the ranging are established,which lays a foundation for the study of the real-time obstacle avoidance of the MAV.Considering that the size and type of the obstacles in the surrounding environment cannot be obtained in advance,it is necessary to design the target obstacle detection and identification method on the basis of the three-dimensional reconstruction of the surrounding environment.In order to avoid the obstacle effectively,a method of obstacle identification based on contour detection is designed.By further judging the size of the contour,the error detection caused by the depth image noise is reduced,which help to identify the obstacle effectively.Based on the analysis of the relative motion between the vehicle and the potential obstacle,and considering the motion characteristics and constraints of MAV,the traditional method of collision cone model is improved and the modified avoidance strategy is designed by the method of motion decomposition.And a visual guidance method is proposed,which takes into account the real-time obstacle avoidance and accuracy.The autonomous obstacle avoidance ability is improved,which provides the foundation for the safe flight of MAV as well.While avoiding obstacle in the flight,the MAV must measure the absolute position and attitude information accurately.However,the traditional inertial/visual odometer combination has the problem of accumulation of errors,especially under the condition of high maneuver of obstacle avoidance flight,which can cause the divergence of inertial error,and the visual odometer has the measurement delay and unknown output error model,which leads to the traditional inertial/visual odometer combination cannot meet the need of the MAV obstacle-avoidance flight.Thus,a binocular vision odometer/inertial integrated navigation method based on modified measurement model is proposed in this paper,using the real-time pose change obtained binocular vision odometer.On the basis of the correlation noise Kalman filter,the on-line error estimation and compensation of inertial devices and inertial navigation system is realized,which improves accuracy of autonomous navigation effectively and provides the absolute pose information for the safe flight of MAV in the complex flight environment.Finally,based on the M100 four rotorcraft and manifold(processing module)as well as the binocular vision sensor module,this thesis designs a visual obstacle-avoidance verification scheme based on binocular camera,and analyzes the proposed visual obstacle-avoidance method of MAV.Then,based on the actual flight test data the effectiveness of the autonomous navigation method is analyzed,which proves that it provides a useful reference for the autonomous obstacle avoidance and safe flight of MAV.