冬季果树林间视觉导航基准线识别算法

Recognition of visual navigation directrix between winter fruit tree row

由冬季树木养护需求出发,结合冬季树木的树干刷白特征,提出一种适用于冬季果树林间行走的机器人导航路径生成算法。首先根据刷白树干特征进行灰度投影截取后续图像处理区域,应用灰度门限法快速分割出图像前景,采用Mid-to-Side扫描方式并结合树干形状几何特征参数判断自底向上提取道路两侧近邻刷白树干,以其内侧角点作为特征点拟合左右边界线,然后求其中线作为导航基准线,最后根据前次检测结果建立动态ROI区域以实现跟踪处理。试验结果表明,算法针对性突出,易于实现,实时性好,在不同时段及不同光照条件下的导航基准线生成准确率高于90%,静态检测处理时平均每帧图像耗时小于274 ms,动态跟踪处理时平均每帧图像耗时156 ms,林间50 m动态连续运行的数据误差波动在可接受范围之内,验证算法具有较好的鲁棒性与抗干扰性,较好满足机器人林间行走视觉导航的基本要求。

Combining with the notable feature of white trunks,the paper suggests a kind of path generation algorithm for robot navigation between winter fruit trees row from the need of maintenance of winter trees.Firstly,part of gray image was extracted for the following steps according to the characteristics of gray horizontal projection.Next,fast segmentation of foreground target using gray threshold method.After that,the trunks nearing two sides of current path were selected through the Mid-to-Side scanning method while white trunk was recognized and judged by the geometric characteristic parameters of the trunk shape.Then fitting two straight boundary lines using inner corner points of white trunks as feature points of each side.Finally the central line was obtained as the navigation guidance line.The results of experiments show that the algorithm is highly targeted,easy to implement,and has good real-time performance,the accuracy of the navigation line is higher than 90%,the average processing time is less than 274 ms per frame and 156 ms per frame for dynamic tracking.The error data fluctuation range of 50 meter dynamic test is within acceptable limits.All above show that the algorithm is verified to be robust and real-time,has good prospects for the promotion of visual navigation and its related applications in forestry environment.