基于机器视觉的旱田多目标直线检测方法

在实际应用中，由于摄像头安装的高度不同或者车辆在地面高低不平的农田内行驶时产生的晃动，都会产生摄像头内出现多作物行的现象。因此根据农田图像的特点，提出了基于已知线的方法判断农作物列数，避免了传统算法只有先确定农作物列数才能提取导航线的弊端。针对农田图像中多列目标检测问题，采用了基于水平线扫描的归类算法，并利用改进的Hough变化快速检测多条定位线。试验结果表明，处理一幅720×480像素彩色图像平均消耗时间为258 ms，98%的图像中所有目标直线都可以检测出来。该算法能够准确提取各种天气环境下农田图像中的列信息，确定多条定位线的方位。

Detection algorithm for crop target multi-lines of the field image based on machine vision

When the installing height of camera was different or when the traffic running in the rough fields, there would be more than one crop lines exist in camera, in practical applications. An image algorithm which could determine all orientation lines of the targets in a crop image was proposed. The image algorithm eliminated the malpractice which need determining the number of orientation lines before extract navigation routes in tradition algorithm. The target regions were obtained by analyzing horizontal scan lines in the crop image, and then the improved Hough transformation was used to extract orientation lines. The average time required for a 720×480 pixels color image was 258 ms and the recognition rate reached 98%. The algorithm is availability in accurately detecting the orientation lines under different weathers.