摘要
针对棉花铺膜播种作业环境复杂,视觉导航路径检测易受光照强度、噪声及划线深度的影响,设计了一种抗干扰能力强、适应性广的视觉导航路径检测算法。构建图像采集系统,实时采集铺膜播种作业图像,基于Y=0.299R+0.587G+0.114B颜色模型对图像进行灰度化处理。针对第1帧图像,首先在浮动扫描区间[M1-k,M2-k](M1取560,M2取639;k=0,k≤560,k++)从第0行开始逐行扫描像素点,提取每行灰度值最小的像素点(或灰度值最小的像素点的列坐标的平均值)作为各行路径提取的候补点,并计算每个扫描区间内的候补点列值的方差Fk;寻求Fk值最小的区间作为第1帧图像的目标区间;在目标区间内使用最小二乘法拟合候补点集群提取初始导航路径;然后,以初始导航路径为中心,左右各扩展U个像素作为扫描区间,提取每行灰度值最小的像素点(或灰度值最小的像素点的列坐标的平均值)作为各行路径提取的候补点;最后,使用最小二乘法拟合导航路径,完成第1帧图像导航路径的提取。从第2帧图像开始,首先确定以前1帧图像导航路径作为当前帧图像扫描区间的中心,左右各扩展U个像素作为扫描区间;然后,从第0行像素开始逐行扫描,并提取灰度值最小的像素点(或灰度值最小的像素点的列坐标的平均值)作为路径提取的候补点,并使用'差异权重法'平滑候补点群;最后,基于最小二乘法拟合导航路径。采集6种工况下铺膜作业视频进行验证试验,结果表明:导航路径检测准确率为100%,平均处理速度为7.020ms/帧,能够稳定、快速地检测导航路径,准确率高,适应性广,抗干扰能力较强,满足棉花铺膜播种作业的实际要求。该检测算法丰富了基于视觉的拖拉机行走路径检测的方法,为实现拖拉机自动驾驶奠定了理论基础。
Aiming at the complex environment of cotton filming and sowing,the visual navigation path detection is susceptible to the influence of light intensity,noise and depth of scribing.A visual navigation path detection algorithm with strong anti-interference ability and wide adaptability is designed.The image acquisition system is constructed,and the image of the filming and seeding operation is collected in real time,and the image is grayscaled based on the Y=0.299*R+0.587*G+0.114*B color model.For the first frame image,firstly,in the floating scan interval[M1-k,M2-k](M1 is 560,M2 is 639;k=0,K≤560,k++),the pixels with the smallest gray value per row(or the average value of the column coordinates of the pixels with the smallest gray value)are scanned row by row,and the variance Fk of the column values of the candidate points in each scan interval is calculated.The minimum Fk value interval is sought as the target interval of the first frame image,and the least square method is used to fit the cluster of candidate points in the target interval to extract the initial navigation path.Then,with the initial navigation path as the center,the left and right extended U pixels are used as scanning intervals,and the minimum gray value of each row(or the average value of column coordinates of the minimum gray value of the pixels)is extracted as candidate points for each line of path extraction.Finally,the least squares method is used to fit the navigation path and extract the navigation path of the first frame image.Starting from the second frame image,the navigation path of the previous frame image is determined as the center of the scanning range of the current frame image,and the left and right U pixels are expanded as the scanning range.Then,the minimum gray value of the pixels(or the average of column coordinates of the minimum gray value of the pixels)is extracted as the candidate points for the path extraction,and the'difference weight method'is used to smooth the candidate points.Finally,navigation path fitting based on least square method.The video of the filming operation under six working conditions was collected for verification test.The test results show that the accuracy of the navigation path detection is 100%,and the average processing speed is 7.020 ms/frame.It can detect the navigation path stably and quickly,with high accuracy and adaptability.It has wide sex and strong anti-interference ability,and can meet the actual requirements of cotton filming and sowing operations.The detection algorithm enriches the vision-based method of tractor walking path detection,and lays a theoretical foundation for the realization of tractor automatic driving.
作者
张雄楚
李景彬
姚庆旺
付威
温宝琴
坎杂
Zhang Xiongchu;Li Jingbin;Yao Qingwang;Fu Wei;Wen Baoqin;Kan Za(College of Mechanical and Electrical Engineering,Shihezi University,Shihezi832000,China;Key Laboratory Agricultureof Xinjiang Production and Construction Groups,Shihezi832003,China)
出处
《农机化研究》
北大核心
2020年第5期33-39,共7页
Journal of Agricultural Mechanization Research
基金
国家重点研发计划项目(2016YFD07011504)
新疆生产建设兵团中青年科技创新领军人才项目(2016BC001)
关键词
棉花铺膜播种
视觉导航
路径检测
最小二乘法
cotton film-spreading and seeding operation
visual navigation
navigation calibration
least square method
