| With the gradual rise of facility agriculture,greenhouse orchard,as a modern production and management mode,is expanding in scale.However,the lack of mechanization of orchards and labor shortage restrict the further development of greenhouse orchards.Therefore,it is increasingly urgent to introduce intelligent equipment into greenhouse orchards to improve production efficiency.It is imperative to use robots to replace manpower to complete production tasks.Taking the standardized greenhouse orchard as the research environment and the wheeled robot equipped with hybrid solid-state lidar and monocular camera as the research carrier,this thesis studies the key technologies such as fruit tree position detection and path extraction,inter row pedestrian detection and identification,and motion control during the operation of greenhouse orchard inspection robot.The main research contents are as follows:(1)The greenhouse orchard inspection robot system is designed and built.Analyze the functional requirements of the greenhouse orchard inspection robot,design the corresponding solutions according to the functional requirements of each part,and build the robot hardware platform.The key environment sensing sensors and other hardware modules are selected and introduced,and the software development environment used in this research is introduced.(2)The problem of fruit tree position detection and path extraction of greenhouse orchard inspection robot is studied.Firstly,the data of the three-dimensional orchard lidar point cloud is preprocessed,the sampling area is divided,and the point cloud is down sampled to reduce the amount of data processing.The ground point cloud with dead branches and leaves is removed by the piecewise least square method.Secondly,the three-dimensional point cloud is projected to the XOY plane under the lidar coordinate system,and the trunk position of fruit trees is calculated from the twodimensional point cloud clustering results to judge the end position of fruit trees.Finally,the target path between fruit tree rows is fitted from the clustering results based on RANSAC algorithm.The experiment results show that the accuracy of the trunk position detection scheme meets the actual needs,and the fitted target path meets the accuracy requirements of the robot’s autonomous operation in the greenhouse orchard.(3)The problem of pedestrian detection and identification of greenhouse orchard inspection robot is studied.Firstly,the orchard pedestrian data set is trained by YOLOv5 s target detection algorithm to obtain the pedestrian detection model.Secondly,aiming at the problem that the pedestrians in front affect the lidar navigation process,the detected pedestrian target frame is analyzed,and the height of the detection frame is used as the active parking threshold of the robot.Finally,the face detection model is trained through YOLOv5 s,and the detected pedestrian target box is used as the input to solve the problem that the proportion of face area in the image is too small,and the pedestrian identity information is verified through Face Net.(4)The function of greenhouse orchard inspection robot system is realized.Firstly,the differential kinematics model of wheeled chassis and the kinematics model of robot linear tracking are established.Secondly,the attitude angle of the robot is limited to solve the problem of path loss,and the motion control of the robot is realized by PID controller.Then,the system control strategy under multiple working conditions such as no pedestrian and pedestrian interference between tree rows is designed.Finally,experiments under multiple working conditions are carried out in the simulated environment to verify the linear autonomous navigation accuracy and overall functional integrity of the greenhouse orchard inspection robot.This thesis has 73 figures,21 tables and 90 references. |
PDF Full Text Request