The Visual Navigation Method Of Intelligent Weeding Robot

It is generally known that China is a big agricultural country,but it is not an agricultural power.The transformation from a big country to a powerful country is a long way to go.It needs our unremitting exploration and efforts.Agricultural automation and intelligent agriculture will be the inevitable development direction in the future.Agriculture can be said to play a mainstay role in the construction of national economy,which affects the national policy and people’s livelihood.Therefore,the agricultural reform is a historical necessity.An efficient,accurate and green modern agriculture is the foundation of people’s happiness.One of the key tasks of farmland operation is weed control.The traditional way of weed control is to spray pesticides on a large area,which will not only cause the waste of pesticides,but also lead to a large amount of pesticide on crops and soil.In view of this situation,this paper will design an intelligent weeding robot to achieve targeted spraying and precise weeding,so the pesticide waste and pollution will be reduced.In addition,the visual navigation algorithm is applied on the robot platform,so that the robot can be adapt to different scenarios required by work,thus completing the corresponding navigation tasks,and realizing the fully autonomous and automatic work.The research of this paper is mainly from the following aspects:First,we assemble the intelligent weeding robot.Here,a four-wheel car driven by two wheels is used as the robot chassis,and a proper mechanical structure design is carried out for this,as well as the layout and connection of hardware will be completed.After that,we design the bottom driver software system,so that the robot can move according to the speed given by upper software.Furthermore,the robot can control the switch of solenoid valve through host computer.The host computer is mainly under the ROS framework,so that messages can be easy to pack and transmit.Secondly,this paper researches the visual navigation algorithm for the platform.A visual navigation method is designed to make the robot travel along the track formed by crops in the field.The main idea is to identify the connected domains of crops,and use the least square method to fit the centroids of the connected domains into a straight line,so that the robot can drive along the ridge and complete the weeding operation.After that,the simulation of spraying while tracking is realized on the real platform,which can verify its feasibility.Then,considering that the robot will be automatically charged to the charging post in the future research,so when the weeding robot leaves the field after the operation and the robot is not in the field,a navigation algorithm based on visual SLAM is designed.In this paper,three different visual SLAM algorithms are tested and compared.Then,select the most suitable sensor and SLAM algorithm for the robot,and optimize and filter the 3D map according to the actual situation.Finally,a reliable 2D obstacle map is generated for the later autonomous path planning.Finally,the global path planning algorithm A-star is studied,and two local path planning algorithms in ROS framework are compared.After that,the global and local algorithms are simulated and verified.Combined with global algorithm and local algorithm,the robot can keep away from the temporary static obstacles under the premise of global path optimization.At last,a combined path planning scheme based on A-star global path planning algorithm and TEB local path planning algorithm is realized,and the effectiveness of the above scheme is verified by simulation.