Research On Path Planning And Object Detection Technology Based On UAV

With the development of science and technology and the advent of artificial intelligence,UAVs have been widely used in various fields,such as high-altitude operations,traffic detection,rescue and disaster relief tasks.This topic focuses on the UAV path planning problem and target detection problem.Solving the path planning problem can improve ability of planning and autonomous flight for UAV.Target detection technology can improve the ability of targets searching for UAVs.Firstly,this thesis focuses on the UAV static 3D path planning problem.Based on the environment,static 3D path planning is divided into global path planning and local path planning in the continuous domain.The thesis mainly focusses on global path planning.Compared with 2D path planning,the computational effort is increased exponentially,so traditional algorithms are no longer applicable.Therefore,this thesis chooses the particle swarm algorithm as the base algorithm for the swarm intelligence algorithm to improve.However,global path planning requires pre-set environmental information,this thesis uses simulated mountainous terrain as the experimental environment.On this basis,simple and complex environments are set up to demonstrate the effectiveness of the algorithm,while random environments are set up to further prove the universality of the algorithm.The particle swarm algorithm has the advantages of fast response time and simple process.While in the late iteration,it has the problem of poor population diversity and easily falling into local optimum.To solve the problem,this thesis proposes an improved particle swarm algorithm,CPSO(Commix PSO).Firstly,the fixed weights are changed to adaptive dynamic inertia weights,which can ensure the search ability in the early stage and improve the search ability in the late iteration.Secondly,the differential evolution algorithm is introduced,combining the selection operation of genetic algorithm and the crossover and variation operation of differential evolution algorithm.It can improve the population diversity of the particle swarm algorithm and avoid falling into local optimum.Finally,the generated paths are smoothed by the three-time B-sample curve operation.To verify the feasibility of CPSO,MATLAB is used to perform simulation experiments for validation.In the same environment,CPSO has significant effect relative to the comparison algorithm.CPSO also outperforms the comparison algorithm in terms of stability and speed in a stochastic environment.Through simulation experiments,the results compared with various algorithms prove that CPSO can be applied to UAV 3D path planning tasks.YOLOv5,a single-stage target detection algorithm,is selected to solve the UAV target detection problem.The detection performance,deployment difficulty and algorithm stability of YOLOv5 have been greatly improved compared to previous generations.However,for embedded devices deployed with small memory,YOLOv5 still has room for improvement in terms of lightweight network structure and improving small target detection capability.To address this issue,an improved YOLOv5 is proposed,which is named C-YOLOv5(Commix YOLOv5).In this thesis,the detection module of YOLOv5 is first reasonably tailored to improve the network lightweighting.Then the SOB-FPN(Scale Out Bi FPN)feature fusion network structure is proposed according to the Bi FPN feature fusion network.It can improve the detection accuracy and avoid the redundancy of network structure.Finally,Coord Att attention mechanism is introduced to improve the overall detection capability and generalization ability of the network model.Finally,the comparison experiment and ablation experiment of C-YOLOv5 were conducted on the Vis Drone2019 test set.Compared with YOLOv3,YOLOv4-tiny and YOLOv5 s,C-Yolov5 has been improved to different degrees.Meanwhile the C-YOLOv5 algorithm performs physical inspection on the UAV platform and successfully completes the task.The experimental results demonstrate that C-YOLOv5 is suitable for UAV target detection tasks.