Field-road trajectory segmentation for agricultural machinery based on direction distribution

Abstract

Field-road segmentation that automatically divides a trajectory into a sequence of field/road segments is an important component in segmentation process for the trajectories of agricultural machinery. A trajectory is a sequence of geospatial coordinates recorded by GNSS receivers during the driving of the machine. The objective of this paper is to develop a field-road segmentation method in the case of the unavailability of field boundary information. The developed method consists of two stages. The first stage uses DBSCAN, a typical clustering algorithm, to do field-road segmentation, and the second stage uses a rule-based inference to correct two types of false segmentation cases from output of DBSCAN-based clustering. Based on the parallel direction distribution that strips in the same field are almost parallel, two inference rules, Field2Road-Cluster and Road2Field-Segment are performed sequentially. Field2Road-Cluster uses the direction distribution difference (parallel in fields vs. not parallel on roads) to correct false field segmentation cases and Road2Field-Segment uses the parallel relationship among strips in the same field to correct false road segmentation cases. The developed method was validated by 60 selected trajectories. The results demonstrated that the rule-based inference achieved an increase of 7.95% in F1 scores, where Field2Road-Cluster and Road2Field-Segment contributed 6.40% and 1.55% increase, respectively.