Target Detection And Motion Planning Of UAV Based On Intelligent Assembly Workshop

With the continuous innovation of industrial technology and manufacturing development,the management of high-end,digitalization and informationization has become an inevitable trend.Unmanned aerial vehicles become the focus of a brand generation of business revolution with their strong interoperability,high flexibility,and excellent maneuverability.At present,several enterprises have step by step opened a new mode of UAVs safety review and quick response so as to make intelligent workshops,cut back labor prices and optimize management mode.With limited hardware system enhancements,efficient target detection and real-time motion planning become the key factors moving UAVs potency.In this paper,Prometheus600 quadrotor UAV is taken as the research object,algorithm optimization and modeling simulation are used as tools to study the target detection accuracy,plan obstacle avoidance path and optimize tracking trajectory.The target detection model in complex environment is explored.The environment model is constructed by workshop 3D information,and the obstacle avoidance path is searched based on intelligent algorithms.And the trajectory is optimized with time-security as constraint.The main research completed in this paper includes:（1）The kinematic model of the quadrotor UAV is discussed,the target detection and motion system are designed.Firstly,the working principle of different flight modes of UAV is analyzed.On the premise of center of mass translation and rotation,the nonlinear model of UAV is constructed according to Newton-Euler law.Then,the observation model of the UAV on the target is established,and therefore the fusion of image info and Li DAR info is projected to realize the 3D positioning of the workshop equipment.Finally,the attitude control based on proportional differential feedback and position control based on hovering and trajectory tracking are proposed to solve the errors caused by system perturbation.（2）An improved Yolov3 target detection technique is proposed.The Yolov3 network has good performance in detecting images,but there are problems such as inaccurate target location,small target easy to miss detection and false detection.Based on this,the initial image is preprocessed by histogram equalization and image filtering.Then,the SPP module is added in front of the Darknet-53 network,the CIo U cross-merge ratio is applied to optimize the loss function,and the K-Means++clustering algorithm is used to calculate the prior frame.The ablation experiments and detection results show that the improved Yolov3 algorithm outperforms other network models in terms of detection accuracy.The detection accuracy rate is 86.9%,the recall rate is 90.5%,the m AP is 92.3%,and the F_1-score is 88.7%.It also has good adaptability to the different target numbers and categories.（3）The mathematical model of UAV path planning is established.Firstly,the basic problem of path planning is expounded,and the constraints of UAV maneuverability are analyzed.The maximum range constraint,obstacle safety distance constraint,energy loss constraint,maximum turning angle constraint,flight height constraint and maximum pitch angle constraint are designed.And a multi-objective and multi-extreme optimization model with constraints is constructed.Then,the obstacle avoidance principle of UAV and obstacles,route and obstacles are introduced.According to the 3D information of workshop,the mathematical model of real workshop is constructed by grid method,which lays a foundation for path planning.（4）The APFA*autonomous obstacle avoidance algorithm based on parameter optimization is studied.Aiming at the defect that the artificial potential field method is easy to fall into the local minimum,the A*algorithm has the problems of discontinuous curvature and too many redundant points.An APFA*algorithm that combines two algorithms and optimizes the parameters is proposed.Firstly,the search step size of the A*algorithm is dynamically adjusted according to the environmental complexity,the potential field value in the artificial potential field is integrated into the cost function of the A*algorithm,and the pseudo-code flow of the hybrid algorithm is designed.Then the key parameters of APFA*algorithm are optimized,the regression prediction model is established and the coupling effect between parameters is analyzed in detail,and the optimal parameter combination is obtained.The simulation results show that the path length planned by APFA*algorithm is 35.8 m in5.64 s,which is about 14.8%and 1%shorter than artificial potential field and A*algorithm.The global search ability,route smoothness,attitude angle and flight height of APFA*algorithm are also better than the other two algorithms.In addition,the reliability and robustness of the proposed algorithm are verified by flight experiments.（5）The trajectory optimization method based on the safety constraint Minimum snap is proposed.Considering that the flight path planned by APFA*algorithm is unstable due to the impact of velocity and acceleration,the optimal trajectory under time-safety constraints is carried out.Firstly,the differential flatness theory is applied to the planned path to reduce the dimension of the state space.Then,the objective function based on Minimum snap is constructed to analyze the smoothness constraints and continuity constraints of higher-order derivatives.Finally,the concept of air corridor is introduced to solve the problem of unsatisfactory trajectory,and the guidance function is added to make the final optimized trajectory close to the original trajectory.Through the simulation and comparison experiments of velocity,acceleration,angular velocity and angular acceleration,the feasibility of the proposed method is verified,and the stability of the flight process is further ensured.