矮化密植枣园收获作业视觉导航路径提取

发布时间：2018-03-23 16:00

  本文选题：机械化　切入点：农作物　出处：《农业工程学报》2017年09期

【摘要】：针对矮化密植枣园环境的复杂性,提出一种基于图像处理的枣园导航基准线生成算法。选用B分量图进行处理,提出"行阈值分割"方法分割树干与背景;根据拍摄场景及视角提出"行间区域"方法剔除行间噪声;通过统计树干与地面交点位置分布区域选取图像五分之二向下区域进行处理;依据树干纵向灰度分布规律,采用浮动窗口灰度垂直投影方法结合形态学开闭运算提取树干区域;基于枣园行间线性分布特征引入"趋势线",而后利用点到直线的距离与设定阈值作比较选取树干与地面的交点;利用交点的位置分布将其归类,并采用最小二乘法原理拟合左右两侧边缘,提取边缘线上各行的几何中心点生成枣园导航基准线。通过对阴天、晴天、顺光、逆光、噪声多元叠加5种条件进行试验,结果表明,该算法具有一定的抗噪性能,单一工况条件导航基准线生成准确率可达83.4%以上,多工况条件准确率为45%。针对5种工况条件的视频检测,结果表明,单一工况条件算法动态检测准确率可达81.3%以上,每帧图像处理平均耗时低于1.7 s,多工况条件检测准确率为42.3%,每帧图像平均耗时1.0 s。该研究可为矮化密植果园实现机器人自主导航作业提供参考。
[Abstract]:In view of the complexity of the environment of dwarf and close planting jujube orchard, an algorithm of generating navigation datum line based on image processing is proposed. B component map is selected to process, and the method of "row threshold segmentation" is proposed to segment tree trunk and background. According to the shooting scene and angle of view, the method of "interline area" is put forward to eliminate interline noise, and the image 2/5 downward region is selected to process by statistics of the distribution region of the intersection of tree trunk and ground, and according to the rule of trunk longitudinal gray distribution, Using floating window gray vertical projection method combined with morphological open and close operation to extract tree trunk area; Based on the characteristics of linear distribution between rows of jujube orchard, the "trend line" is introduced, and then the distance from point to line is compared with the set threshold to select the intersection point between tree trunk and ground, and the intersection point is classified by the position distribution of intersection point. The principle of least square method is used to fit the left and right edges, and the geometric center points of each line on the edge line are extracted to generate the navigation datum of jujube garden. The experiments are carried out under five conditions: overcast, sunny, smooth, inverse and noise. The results show that the algorithm has a certain anti-noise performance, the accuracy rate of generating navigation datum under a single working condition can reach more than 83.4%, and the accuracy rate of multi-condition condition is 455.The video detection of five operating conditions shows that, The accuracy of dynamic detection of single working condition algorithm can reach more than 81.3%. The average time of image processing per frame is less than 1.7 s, the accuracy of multi-condition detection is 42.3 and the average time of each frame is 1.0 s.This study can provide a reference for autonomous navigation of robot in dwarf and dense orchard.
【作者单位】： 石河子大学信息科学与技术学院;石河子大学机械电气工程学院;
【基金】：国家重点研发计划课题(2016YFD07011504) 兵团中青年科技创新领军人才(2016BC001) 石河子大学杰出青年科技人才培育计划(2014ZRKXJQ04)
【分类号】：TP242;TP391.41
，

本文编号：1654160