Moving Target Recognition And Tracking For Quadrotor UAV

Air-ground cooperation technology of UAV and UGA has been widely used in military,civilian,disaster rescue and other fields.At the same time,moving target recognition and tracking are the application fundamentals and key technologies.On the other hand,quadrotor UAV with its simple structure and flexible mobility has become popular in the field of UAV research.According to Mission 7 of International Aerial Robotics Competition(IARC),this thesis researches and realizes the system of moving target recognition and tracking for quadrotor UAV,which is of high scientific and application value.Firstly,data of optical flow and inertial measurement unit(IMU)are fused with the Bayesian filter to estimate the state of the UAV.The posture information from IMU is used to transform the perspective of the image while the actual distance is calculated by using scale transformation.Besides,fisheye lens is calibrated to expand UAV field of view and the agent can be identified by HSV color space image segmentation and corrosion expansion method.The direction of motion target is estimated by background compensation optical flow.Experiment results show that the algorithm can meet the requirement of UAV location and ground target recognition.Secondly,in order to get state transition and the state prediction model of quadrotor UAV,state and motion model are established.State observer model is also established according to characteristics of IMU and optical flow.Navigation control system has communication,calculation,display and other functions.Controlled quantity is calculated by PID controller based on EKF,and GUI can transmit and display message.According to mission of air-ground cooperation in IARC,finite state machine model is designed so that the UAV can be reasonably converted in the search target,target tracking,obstacle avoidance and other models.Thirdly,a solution of obstacle avoidance for AR.Drone based on ultrasonic sensor is presented.In order to get an optimal path from start to finish point without collision,traditional Artificial Potential Field(APF)avoidance algorithm is improved,which uses Cross-Entropy(CE)optimization method.Then the avoidance strategies under different circumstances of both single obstacle and multi-obstacles are discussed.In addition,the fractional potential field(FPF)is deduced,and the experiment results shows it is more flexible,diverse and suitable for complex environment.Finally,the hardware and software of AR.Drone are improved.The recognition and control system on ground station are developed based on Robot Operating System(ROS).The obstacle avoidance algorithm is tested with AR.Drone toolbox in MATLAB.In order to verify the feasibility of system,a simulation system is established based on Gazebo.Besides,an experiment is also tried in real environment.Both simulation and reality experiment results show that the system have the capability to track randomly moving objects and interact physically with them.