基于双目视觉的作物苗期障碍物三维信息检测方法

Three-dimensional Information Detection Method for Crop Seedling Obstacles Based on Binocular Vision

农业智能装备在实际农田环境中行进或作业的过程中需要感知多变环境下的各种障碍物。为此,基于双目视觉,开展了作物苗期农田障碍物三维信息检测方法研究,提出了一种基于特征的障碍物检测算法。首先,利用边缘检测算法去除天空背景,提取出障碍物潜在区域的上边界线,利用超绿特征颜色变换去除绿色作物苗期农田背景,提取下边界线;然后,通过阈值分割算法提取障碍物目标区域;最后,通过重心特征点立体匹配来获取视差值,结合MatLab标定获取的相机内外参数进行三维重建,计算障碍物的距离、宽度和高度三维信息。田间试验结果表明:该算法可以正确提取出障碍物目标区域,障碍物距离、宽度和高度检测的平均相对误差分别为4.7%、5.79%和1.78%,能够满足农业智能装备田间障碍物检测的需求,具有较好的可靠性。

Agricultural intelligent equipment needs to perceive various obstacles in a changing environment during the process of traveling or working in an actual farmland environment.Based on binocular vision,this paper studies the three-dimensional information detection algorithm of farmland obstacles in crop seedlings,and proposes a feature-based obstacle detection algorithm.Firstly,the upper boundary line of the potential area of the obstacle is extracted using the edge detection algorithm which is used to remove the sky background and the lower boundary line is extracted using the green crop color change which is used to remove the green crop seedling farmland background.Then the obstacle target area is extracted by the threshold segmentation algorithm.Finally,the disparity was obtained by stereo matching of barycenter feature points,and three-dimensional reconstruction was carried out based on internal and external camera parameters obtained by Matlab calibration,and distance,width and height of obstacle were calculated.The field experiment results show that the algorithm can correctly extract the obstacle target area,and the average relative error of obstacle distance,width and height detection are 4.7%,5.79%and 1.78%,respectively,which can meet the needs of agricultural intelligent equipment field obstacle detection.