Research On Agriculture Mobile Robot Tree Trunk Detection And Navigation In Orchard Based On Laser Radar

China is a large agricultural country,to reduce labor intensity and production costs of the agricultural operations,using agricultural mechanization is imperative,and agricultural mobile robot is in so great need to become a research hotspot.Currently agricultural mobile robot has been made many important advances in farmland areas,while orchard environment is more complex and it is relatively more difficult to research agricultural mobile robot which can be applied in orchards.For various work in orchard production management process,first,the agricultural mobile robot have to understand the orchard environment and identify the location of the trees,then,walking in tree rows safely and accurately.In this paper,we use laser radar as navigation detection sensor,designed a tree detection method and achieved the agricultural mobile robot autonomous navigation.The main researches were as follows:(1)Using DBSCAN clustering algorithm to gather and classification the data from the environment collected by laser radar.DBSCAN clustering algorithm is very sensitive to the input value of two global thresholds:the cluster radius and the quantity of points within the class.We proposed dynamic adaptive thresholds according to points distance to improve this kind of sensitive,so the clustering algorithm can detect various diameter trees in distances.We also improved the search method of the clustering algorithm to make it run faster.(2)Agricultural mobile robot bump while walking on uneven road,this may cause the laser radar scanning plane declination to hit the ground or tilting upward to hit the branches,these points will be clustered to confuse the robot;In addition,the laser radar may miss detect the fine tree at a distance.Against these two cases,we proposed to use odometer dead reckoning model to calculate temp robot pose in current frame (?),then used (?) and robot pose in previous frame to track and match vicinity classes in two frames,then use the matched class to correct robot pose in current frame (?),then use the correct (?) and positions of the undetermined class in robot coordinates in different frames to inference and identify classed at a distance.(3)Use tree detection results to generate navigation line and control value to guide agricultural mobile robot walking in tree rows,use vc++ programming language complete the design of overall system.The results showed that the algorithm can identify the data class in 3 frames and recognize the fine tree,and successfully navigate in the tree row with information of detected trees.