基于逆透视变换的条播作物早期作物行识别  

Identification of early crop row for drillcrops based on reverse perspective transformation

作　　者：赵学观[1,2] 马伟 高原源[1] 臧云飞 何义川 王秀 ZHAO Xueguan;MA Wei;GAO Yuanyuan;ZANG Yunfei;HE Yichuan;WANG Xiu(Beijing Research Center for Intelligent Equipment Technology in Agricultural,Beijing 100097,China;National Engineering Research Center Information Technology in Agricultural,Beijing 100097,China;School of Agricultural Engineering and Food Science,Shandong University of Technology,Zibo,Shandong 255000,China;Xinjiang Agriculture Academy,Shihezi,Xinjiang 832000,China)

机构地区：[1]北京农业智能装备技术研究中心,北京100097 [2]国家农业智能装备技术研究中心,北京100097 [3]山东理工大学农业工程与食品科学学院,山东淄博255000 [4]新疆农垦科学院,新疆石河子832000

出　　处：《江苏大学学报（自然科学版）》2019年第6期668-675,共8页Journal of Jiangsu University：Natural Science Edition

基　　金：国家"十三五"重点研发项目(2016YFD020060403)

摘　　要：根据条播作物田间作业的对行要求,提出了一种新的作物行检测方法.首先通过选取透视图像中底部区域进行垂直投影,获得透视图像中作物行的边缘点并进行逆透视变换,然后基于逆透视变换算法消除图像几何失真,利用骨化算法求取逆透视图像中作物行的骨架线交点集,在逆透视变换图像中根据边缘点对骨架线交点集进行划分,最后对交点集分类后的作物行进行拟合,以获取不同的作物行.通过对300幅不同生长条件下的小麦作物行图像进行识别试验,其行识别拟合结果表明:作物行拟合的平均误差为2.136 7°,标准差为1.024 3°,平均耗时为0.364 7 s,能够满足实时工作要求.To meet the requirements of field working along the drill crops, a new crop row detection method was proposed. The camera was installed in front of the tractor to obtain the relative space position of farm implements or crop rows. To improve the recognition accuracy of crop rows, the inverse perspective transformation algorithm was used to eliminate the image geometric distortion, and the ossification algorithm was used to obtain the intersection of the skeleton lines of crop rows in the reverse perspective image. The bottom area of the perspective image was selected for vertical projection, and the edge points of the crop rows in the perspective image were obtained. The inverse perspective transformation was carried out, and the intersection points of skeleton lines were divided in the inverse perspective image according to the edge points. The results of identifying and fitting 300 wheat rows with different growth conditions show that the average error of crop row fitting is 2.136 7 degree with standard deviation of 1.024 3 degree and average time consuming of 0.364 7 s, which can meet the real-time requirements.

关 键 词：作物行识别 导航 逆透视变换 机器视觉 条播作物 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]