Vineyard Skeletonization for Autonomous Robot Navigation

Prior knowledge of possible routes is undoubtedly an added value for autonomous navigation in irregular agricultural terrains. This information is particularly important when it involves the navigation of a monitoring robot, which necessarily carries a wide range of expensive sensors and when the vineyard presents a non-uniform configuration and extends over a very highly uneven terrain. In such case, a small navigation positional error can result in a large vertical deviation and consequently, a serious fall, which may damage or even destroy the robot. This article presents an automated way of deriving possible routes in this kind of terrain using three curve-skeleton algorithms for the 3D surfaces of the vineyard where the robot may navigate. The skeleton curves and real trajectory were represented in a graphical user OpenGL application developed for this purpose. A thinning, a geometric and a distance field algorithm were used for this study. The skeleton curves were compared with a real navigation path made by an expert when driving a tractor while spraying of the vineyard. In order to meet expert recommendations, the thinning algorithm was validated as the most suitable to achieve the aim of the study as it minimizes the quadratic average distance function applied to the skeleton points and the real trajectory. The limits of the most suitable curve-skeleton will be used as decision making points to establish navigation criteria for next step path planning.