| In recent years,UAVs have been widely used in the fields of military,agriculture,forestry,and mining engineering.The research on the flight control of UAVs is also a focus of attention of every country and relevant scientific research organizations,such as: in terms of photography and imaging,UAVs can be used for aerial filming and photography,obtaining news data and geological information of mines,and aerial surveying and mapping,etc.;in terms of agriculture,UAVs can be used for pesticide spraying,and monitoring of agricultural production,etc.;in terms of mine information engineering,UAVs can deal with various complex geological structures of mine galleries,harsh and dangerous environments of mining,and conduct inspections of mine galleries,gas explosion and equipment,etc.Because of simple operation,quadrotor UAVs can achieve certain wind resistance and load-carrying capacity,and has the characteristics of being able to hover,the flight control and navigation of quadrotor UAVs has also become a research hotspot in the field of UAVs in complex environments.The combination of GNSS and inertial navigation has been fully studied by a large number of researchers,but it is limited when used in areas where the GNSS signal effect is not obvious.In view of such reason and application scenario,visionbased sensors with the integration of multiple technologies are loaded into UAVs,so that they can use vision technology to carry out many special flight missions such as autonomous navigation and positioning,obstacle avoidance and autonomous flight,target tracking and map construction.Based on such requirements and research target positioning,the quadrotor UAV was taken as the research object in the Paper to estimate the state of the UAV based on computer vision and complete the intelligent flight control after the surrounding environment was sensed.The main research results were as follows:(1)The rotor aerodynamics model and system were modeled for the prototype with a variety of sensor kits,starting from the flight control dynamic modeling of the quadrotor UAV.“Cross-type” and “X-type” quadrotor UAV test platforms were designed and built.By improving the isolation strategy and the airborne flight control circuit system,the “improved “X-type” quadrotor UAV test platform was designed and built.A distributed layered flight control strategy was proposed: With the consideration of the time scale separation between the attitude dynamic model and the positioning control,the dynamic model could be represented by nested and layered height and yaw control,forward position and pitch control,and lateral position and roll angle control.In the upper control,the position control was used to represent the control of the outside loop.In the lower control,the attitude control was realized by the airborne embedded flight control self-driving device.Through hovering flight test verification,the hovering effect,robustness against external disturbance,sensor load,vibration isolation efficiency,etc.of “X-type” quadrotor UAV were compared with those of “improved“X-type” quadrotor UAV.At the same time,the distributed layered flight control strategy was verified.The test results showed that the distributed layered flight control strategy proposed in the paper can guarantee the sound stability of the quadrotor UAV.At the same time,compared with the “X-type” quadrotor UAV,the dynamic performance,anti-disturbance,sensor load and vibration isolation benefits of the“improved X-type” quadrotor UAV have been significantly improved.(2)The optical flow sensor was equipped to meet the stable flight control requirements of the quadrotor UAV in flight,using the optical flow algorithm,that is,based on the optical flow method to carry out mathematical modeling,calculate the angular velocity and attitude angle,to obtain the motion velocity of the quadrotor UAV in the camera coordinate system.Hence,the pose of the quadrotor UAV was estimated and the hovering aircraft was controlled.Firstly,based on the theory of visual image processing,the velocity position estimation method of the optical flow method was analyzed and the mathematical modeling of flight velocity and optical flow was completed.Secondly,the advantages and disadvantages of common image interpolation optical flow algorithm and Lucas-Kanade optical flow algorithm were analyzed,on which basis an improved optical flow algorithm of LK fusion feature matching was proposed.Median filtering was added to the algorithm to remove some noise.Feature matching was used to complete the optical flow rough calculation,and then LK was used to carry out the iterative optical flow calculation.By setting the proportion coefficient of adjacent frames and the threshold value,the outliers were removed and the robustness of the algorithm was improved in the weak light environment and the strobe light environment.Finally,the performance of the algorithm was verified and compared through the stable hover test of the quadrotor UAV.(3)In combination with the SVO algorithm,the initial relative motion between adjacent image frames collected by a monocular camera was estimated.Then the optimal solution of the relative pose was calculated and solved by the iterative method.In this way,it could be made full use of without dense data volumes or descriptors.The characteristics of real-time could be achieved in the embedded platform with low performance and low power consumption.By combining the IMU(inertial measurement unit)in a loose-coupling way,using the IMU’s high real-time sensitivity,and adopting the pre-integration model,it was possible to estimate the relative motion in a short period.IMU was used to estimate the relative motion to get the initial value,which can be used as an input for the iterative solution of the SVO sparse direct method.In this way,a better initial value could be obtained.At the same time,the state value of IMU could be updated by combining the calculation results of SVO,which solved the problem of drift in acceleration and angular velocity measurement bias caused by IMU’s attributes,thus improving the accuracy,stability,and robustness of UAV pose estimation in rapid motion.(4)A binocular camera was used to obtain the image data collected by the left and right cameras.The effect brought by the optical distortion was removed by stereo calibration of the camera.After the solutions of the camera distortion parameters,external parameters,and internal parameter matrix were completed,GPU acceleration was used to carry out stereo correction.Firstly,a reverse mapping relation table of pixel points was established according to the distortion parameters as well as the internal and external parameters of left and right cameras.The pixel point coordinates were read into the GPU,and the distortion pixel coordinates were searched and mapped to nondistortion coordinates for stereo correction,by which the real-time requirements could be met.Secondly,the disparity map was obtained by pixel matching method to complete stereo matching,and the depth information was obtained by triangular similarity principle.Finally,the convolutional neural network algorithm was applied to classify all kinds of the collected mine pit road image data,and different flight control strategies were designed according to the divided mine pit road environment,such as gallery,corners,stairs,obstacles,etc.The algorithm verification was completed by obstacle avoidance and collision-free flight tests. |
PDF Full Text Request