| The objective of this thesis is to enhance the uninterrupted operational performance and field navigation accuracy of a safflower picking robot and improve its ability to navigate through unstructured safflower gardens.Simultaneously,it seeks to avoid inflicting damage upon safflower plants during the picking phase,ensuring the safety and stability of the safflower picking robot in the field.This thesis focuses on the field environments of safflower gardens to achieve these objectives.It analyzes the planting patterns of safflower in Xinjiang and the structural characteristics of cross-row picking robots.This thesis introduces a field navigation system for safflower picking robots based on the fusion of multiple data sources.This navigation system integrates positioning data from DGNSS,divider pressure monitoring devices,ZED2 stereo cameras,IMUs,and wheel speed sensors to acquire the position information of the robot and information about the safflower garden environment.The system employs path planning and fusion positioning techniques,interrow navigation methods,and headland turning-based navigation methods to achieve autonomous navigation for safflower picking robots in the field.The research content of this thesis primarily includes the following points.(1)The research mainly entails the design and implementation of a field navigation system for safflower picking robots.This thesis elucidates the overall structure and navigation principles of the field navigation system and details the designs of an interrow navigation subsystem and a headland turning-based navigation subsystem,while also conducting research on obstacle avoidance strategies.The thesis also provides a brief overview of the selection process and technical parameters of the key components in the navigation system.Furthermore,it involves the development of navigation control programs,visual headland data acquisition and processing programs,and relevant communication protocols.(2)This thesis proposes a full-coverage path planning method that combines shuttle rows and Bezier curves,making it suitable for field navigation of safflower picking robots.The fusion positioning algorithm of DGNSS,a wheel speedometer and an IMU based on extended Kalman fusion technology are proposed to solve the low positioning accuracy and poor stability of a single sensor in a farmland environment.(3)This study analyzes the distributions of safflower stem diameters and their bending characteristics.It develops an interrow navigation method for safflower picking robots using a front-wheel differential drive.The research covers trapezoidal velocity planning and a double-loop,second-order fuzzy PID control algorithm,utilizing lateral and heading deviations as inputs.To address the local path curvature phenomenon in safflower rows,a divider pressure navigation algorithm and a Grubbs-based filtering strategy are proposed.In divider pressure navigation tests,when safflower rows exhibit significant curvature,the robot maintains an average path tracking error below 70 mm.The robot can adjust its position to prevent damage from being inflicted upon safflower plants during interrow navigation.(4)This thesis presents headland turning and row-oriented methods,as well as control strategies for safflower picking robots.Headland turning primarily involves segmenting the headland from the background features using the HSV-fixed threshold segmentation method to obtain headland boundary information.A Laplacian sharpening filter is employed to enhance the headland boundary features.The RANSAC algorithm is used to fit the headland boundary line,and the depth distance of the headland boundary line is obtained in conjunction with the depth positioning method of the ZED2 stereo camera.A safflower picking robot with a headland turning trajectory is established based on a third-order Bezier curve,and a path tracking model for headland turning is developed using this trajectory.The row orientation process in the headland primarily involves dividing the ROI and employing the improved supergreen 2G-R-B algorithm and OTSU algorithm to obtain headland row boundary features.The RANSAC algorithm is utilized to fit the headland row boundary lines,and the centerlines of the headland rows are extracted through a headland row-oriented deviation model.Yaw information is obtained for the safflower picking robot by fusing machine vision and divider pressure positioning data using weighted averaging and the maximum value method.(5)To assess the effectiveness,robustness,and accuracy of the proposed navigation system for safflower picking robots and to evaluate the performance metrics produced by fusion positioning methods,interrow navigation methods,and headland turning navigation methods,a series of tests were conducted.These tests included multisensor fusion positioning tests,interrow navigation tests,headland turning tests,and headland row-oriented tests.The multisensor fusion positioning test demonstrated that the fusion positioning method based on an EKF and incorporating DGNSS,an IMU,and a wheel speedometer effectively resolved brief signal disruptions caused by obstructions.It exhibited high positioning accuracy and stability.During the interrow navigation test,a speed comparison test determined that the safflower picking robot maintained a steadystate error of 35 mm when traveling at a speed of 1 km/h,satisfying the performance requirements of interrow navigation.During constant speed tracking tests conducted under four different initial poses,the steady-state error remained under 42 mm,indicating the ability of the robot to adjust for unusual orientations.The robot also exhibited high path tracking accuracy and robustness,making it suitable for navigating through narrow safflower rows.In the headland turning test,during the headland recognition and positioning test,the safflower picking robot exhibited high accuracy in terms of detecting headland boundary lines,with an accuracy rate exceeding93%.The angular deviations of the headland boundary lines were less than 1.5°,and the depth value error was under 50 mm,enabling precise headland boundary line detection.In the headland turning path tracking test,the safflower picking robot achieved excellent path tracking accuracy and robustness at different turning speeds,with average lateral deviations of 37 mm at a turning speed of 0.5 km/h and 53 mm at a speed of 1km/h.This demonstrated the ability of the robot to satisfy the field requirements for performing headland turning in safflower harvesting scenarios.In the headland row-oriented test,during the headland row-oriented recognition and positioning test,which involved identifying safflower headland rows during the harvest season,the RANSAC method outperformed the Hough transform and least-squares methods in terms of row fitting,achieving an optimal recognition accuracy of 90% with an average angular error of 2.7°.Through a comparative row orientation test conducted among different navigation methods,when combining visual and divider pressure navigation,the plant damage rate was as low as 0.33%,which was significantly lower than that induced when using either machine vision or divider pressure navigation independently.This test demonstrated that the fusion of visual and divider pressure navigation for headland row orientation performs exceptionally well,effectively reducing the damage inflicted upon safflower plants by the safflower picking robot.In the headland roworiented path tracking test,even in cases with abnormal headland row recognition effects or poor vehicle postures,the fusion system was capable of making yaw adjustments.The steady-state error induced during headland row orientation was less than 30 mm,indicating that the proposed system possesses strong adaptability in the field,high precision,and high reliability when performing headland row orientation during safflower harvesting,satisfying the requirements of this specific task. |
