Research on trunk recognition and obstacle avoidance technology of woodland mower based on machine vision

<bold>Objective</bold> In view of the forest environment of pecan forest with tall trees and different spacing，the accurate identification and positioning of the trunk and real-time obstacle avoidance of the lawn mower are the prerequisites of ensuring efficient weeding. <bold>Method</bold> In this study，real-time image information of the forest floor was collected by binocular camera.Based on YOLOv8n lightweight trunk root detection algorithm and third-order Bezier curve obstacle avoidance path method，fast identification，accurate positioning and real-time obstacle avoidance are realized.Firstly，according to the RGB and depth images derived from the binocular camera，the YOLOV8n detection model was used to identify and locate the tree trunk roots；secondly，the position of the tree trunk roots located under the camera coordinate system wasused to generate the optimal Bezier curve obstacle avoidance path through the anti-collision and curvature constraints；finally，the obstacle avoidance path under the camera coordinate system is transformed into the latitude/longitude coordinate system to carry out the final autonomous obstacle avoidance work. <bold>Result</bold> The results of the trunk root detection show that the recognition accuracy of the trunk root is 89.9%，there is basically no missed detection；the results of the trunk root positioning accuracy test show that the average relative error rate of the measured longitudinal distance and lateral distance is 1.07% and 3.46% respectively；the results of the obstacle avoidance simulation show that the closest distance between the lawnmower and the tree in the process of obstacle avoidance is 6.31 cm，and the maximum transverse error of the tracking of the obstacle avoidance path is 2.01 cm，the average lateral error is 0.85 cm；the results of the mower obstacle avoidance test show that the closest distance between the mower and the tree is 15.26 cm，the maximum lateral error and the average lateral error are 6.23 cm and 2.71 cm，respectively.At the end of the obstacle avoidance process，it can accurately return to the global path. <bold>Conclusion</bold> When this method is applied to complex woodland mower，it can quickly identify and accurately locate the position of tree trunk roots，and plan the obstacle avoidance trajectory to avoid the obstacle tree operation in a short period of time.The results provide a reference for the unmanned operation of woodland mowing machinery.