| Autonomous navigation technology is the key technology for agricultural robots to achieve autonomous operation.With development of navigation technology and improvement of machine vision technology,the autonomous operation of agricultural robots in orchards has gradually become a reality.In this study,the crawler fruit collector was used as a mobile platform.The research on inter-row visual navigation and trunk tracking control technology of collector was carried out.The inter-row navigation information acquisition method based on the fusion of sky and soil navigation curves was proposed.The three-dimensional information acquisition method of fruit tree trunk was studied,and the inter-row navigation and trunk tracking control system was constructed based on steering characteristics of collector.Finally,the inter-row navigation and trunk tracking were tested and analyzed in orchard.The main research contents are as follows:(1)The terrain of mountainous and hilly orchards and the planting agronomy of fruit trees were analyzed.A pitch real-time search device with multiple pitch angles was designed to trigger the acquisition of images.The device can realize image acquisition based on 60° up and down the horizontal line.A frame is collected every 15°,and a total of 8 frames of images are collected in same period.The HSV color space was used to preliminarily segment the soil,canopy and sky scenes.The optimal segmentation threshold was calculated by OTSU for binarization.The soil,canopy and sky scenes were accurately extracted by maximum connected domain segmentation algorithm.The navigation curve was fitted by bilinear least squares.The final navigation path was obtained by dynamic weighted filtering and Kalman filter algorithm fusion.After testing,the navigation path extraction accuracy of the algorithm was about 86.3%,and the algorithm takes 60 ms.(2)The 3D information acquisition method of fruit tree trunk based on semantic segmentation and point cloud processing fusion was studied.The U-Net network was used to train the semantic features of fruit tree trunk.After model training,the recognition accuracy and recall rate of testing set were 93.88% and 98.92% respectively.The semantic segmentation of trunk area of fruit tree was carried out in two-dimensional image through U-Net network.Combined with coordinate mapping relationship between twodimensional image and three-dimensional point cloud,the trunk area of fruit tree was accurately segmented in point cloud image.And the voxel filtering algorithm was combined for downsampling to reduce the number of point cloud data.The RANSAC cylindrical fitting algorithm was used to segment the trunk point cloud area of the fruit tree.Finally,based on PCA principal component analysis,the trunk axis extremum extraction was performed to achieve high-precision extraction of the trunk threedimensional coordinates.(3)The slip compensation mechanism based on steering characteristics of the collector’s crawler was studied.The mapping relationship between actual steering radius,actual steering angular velocity and theoretical steering radius,theoretical steering angular velocity during the navigation process of the crawler was constructed.And the mapping relationship was used as the slip compensation coefficient during the actual navigation.Through the inter-row navigation deviation acquisition method,the steering radius of the collector was calculated by combining the slip compensation coefficient,and the inter-row navigation of the orchard was realized by combining the preview point tracking algorithm.Through the trunk tracking deviation acquisition method,the deviation information of the three-dimensional coordinates of the fruit tree trunk relative to the reference point of the collector was calculated in real time.And the accurate tracking of trunk by collector was realized in combination with path planning.The navigation software was developed on C++/QT platform combined with machine vision library.And the navigation control software of crawler fruit collector was designed.The steering parameter identification and differential steering kinematics analysis of the collector were combined to realize the autonomous and stable operation of the collector in orchard environment.(4)The orchard navigation and trunk tracking control system of crawler fruit collector was developed.This system was mainly composed of upper computer notebook,PCB lower computer controller,pitch real-time search device and gear shift device.The inter-row navigation and trunk tracking experiments were carried out in field orchard.Among them,when the collector was traveling at a speed of 0.6 m/s in a weed-free environment,the maximum inter-row navigation deviation was 0.14 m,and the average of deviation was 0.086 m,and the root mean square error RMSE was 30 mm.The interrow navigation was stable and reliable during the test.The maximum inter-row navigation deviation in weedy environment was 0.24 m,and the average of deviation was 0.13 m,and the root mean square error RMSE was 55 mm.The environmental test also verifies the robustness of navigation system.The accuracy meets the navigation requirements in the orchard environment.On the basis of inter-row navigation,the maximum lateral offset of trunk tracking experiment was 0.19 m,and the average of lateral offset was 0.16 m,and the root mean square error RMSE was 27 mm.The maximum yaw angle was 23.7°,and the average of yaw angle was 17.8 m,and the root mean square error RMSE was 3.6.The main reason is that the planting agronomy cannot achieve tracking operation in a limited space.Therefore,the overall deviation tracking accuracy was not high,but the accuracy meets the needs of autonomous tracking and collection operations in orchard environment.The inter-row navigation and trunk tracking algorithm described in this paper can fully realize the autonomous operation of agricultural robots in orchard. |
PDF Full Text Request