Development of a robot with 3D perception for accurate row following in vineyard

Abstract

In this paper, a localization and navigation guidance algorithm based on the tree row in orchards for agricultural robots was proposed. In this algorithm, the tree rows formed by the trunks of parallel planting grape trees were used as auxiliary information. Together with the relative position provided by IMU, odometer and 3D LiDAR, the proposed algorithm was used to calculate the position and orientation of the moving robot. Firstly, the coordinates of the robot were obtained by the vehicle positioning system, and then the orientation of the tree and the posture of the robot relative to the tree row were determined through the 3D LiDAR point cloud, to ensure that the robot can parallel walk along the tree row and maintain a specific distance. Finally, the odometer, IMU data and LiDAR point cloud image were deeply fused to determine the accurate position and orientation to complete the path planning in the orchards. Experimental results show that the algorithm can effectively improve the positioning accuracy of the robot walking in the orchard and ensure that the robot can walk along the tree row without hitting the trees.

Accurate Detection and Precision Spraying of Corn and Weeds Using the Improved YOLOv5 Model

“…Although it has a good weeding effect, a large number of spraying pesticides not only pollute the environment but also affect the growth and development of crops, and are easy to produce pesticide residues [3].Precision spraying is an effective way to improve the utilization rate of pesticides, and the accurate identification and location of the target is the premise to achieve precision pesticide application technology. At the moment, machine learning [4]- [7], ultrasonic sensor [8]- [10], laser radar [11], spectrum analysis [12]- [14], and spectral imaging [15]- [17] are the most widely utilized target accurate identification and location technologies. Spectral analysis technology is susceptible to outdoor light and has the disadvantages of a large amount of spatial data and difficult analysis.…”

A panoramic driving perception fusion algorithm based on multi-task learning

“…To improve the processing effect on point cloud data and reduce the complexity of subsequent point cloud data processing, PassThrough filtering [ 33 ] is introduced as a preprocessing step for laser point cloud. PassThrough filtering is a preprocessing method used for laser point cloud processing, aiming to remove invalid data in the vertical or horizontal direction, and is suitable for tasks of extracting specific attribute areas in the point cloud.…”

“…To improve the processing effect on point cloud data and reduce the complexity of subsequent point cloud data processing, PassThrough filtering [33] is introduced as a preprocessing step…”

Setting Method of Downsampling Factor and Grid Factor for NDT Relocation Algorithm in Dynamic Environment

“…Furthermore, the amount of data acquired by a LiDAR system is less than in the case of a camera. Hence the LiDAR sensor is used frequently for positioning and object detection [22]. However, most LiDAR-based localization solutions with prior point-cloud maps [23]- [26] assume that the road scenes are relatively constant, while new constructions, roadside vegetation, partial occlusions by changing objects may severely compromise robustness.…”