采棉机视觉导航路线图像检测方法

Detection for navigation route for cotton harvester based on machine vision

采棉机视觉导航路线的检测是实现采棉机自主导航的重要前提。收获时期的棉田中主要有棉秆、棉花、棉桃、棉叶、杂草及田外区域等多种目标特征,并且机械采收会造成已收获区的棉枝上残留少许棉花,对采棉机田间作业路径的检测造成困难。该文主要通过分析已收获区、未收获区、田外区域、田端的不同颜色特征,对采棉机田间作业路径、棉田边缘、田端等的检测算法。首先采用3B-R-G、|R-G|、|R-B|及|G-B|等颜色分量差的方法,针对棉田内、外等不同区域的目标特征分别进行提取,并以设定的步长进行移动平滑处理,而后基于最低波谷点向未收获区方向寻找波峰上升临界点以及与前一帧直线检测结果相关联等方法,确定直线变换的候补点群,最后基于过已知点Hough变换提取导航直线;试验证明,该算法提取的直线能够准确贴合已收获区与未收获区分界、田侧边缘等,处理时间平均为56.10ms/帧,满足采棉机田间实际生产作业的需求。该研究可为小麦、玉米等其他作物机械化收获时视觉导航路线的检测提供参考。

Auto-navigation has a great significance in increasing the operating quality and production efficiency of agriculture machinery, such as improving the working environment and security situation for workers, reducing the labor intensity, etc. The vision navigation has many technical advantages that it can adapt to the complicated field of the operating environment, has wide detection range and has rich and complete information. It is the research focus in the field of agriculture machinery auto-navigation. How to extract routes fast, accurately, and effectively in the natural environment is the key in vision navigation. The vision navigation routes’ detect of a Cotton-picker is the main premise to achieve its auto-navigation. There are many difficulties in detecting the operation routes of a cotton-picker in the field. For example, during harvest, there are many kinds of target features like stalks, cotton, bolls, leaves, weeds in the cotton field and its outside region. In addition, a little cotton may be left on the stalks in the harvested region when we use machine to pick. This paper puts forward the detection algorithms of the operation routes of a cotton-picker, the edge and the end of the cotton field by analyzing the different color features of the harvested region, the un-harvested region, the outside region, and the end of the field. First, we used the difference of color components, such as 3B-R-G, ｜R-G｜, ｜R-B｜ and ｜G-B｜ to extract the target features of the inner and outside of the cotton field respectively, and smooth the image using the moving average method by the set length. Then by finding the rose critical point of the crest that is based on the lowest trough point to the un-harvested region, and connecting with the line detect result of the previous frame, we determine the alternate point group of a straight line transform. At last, we extracted the navigation routes based on Passing a Known Point Hough Transform （PKPHT）. The test proves that the extracted line by this algorithm can match the harvested region, the un-harvested region and the edge of the field accurately. The processing time is 56.10 ms per frame, which can meet the demand of the real production of a cotton-picker in the field. This research can provide the reference to the vision navigation routes’ detection of wheat, corn and many other crops when we harvest them by machine.