Integration of an Artificial Perception System for Identification of Live Flammable Material in Forestry Robotics

摘要

In this article, we present the integration of an artificial perception pipeline for a heavy-duty forestry robot for automatic identification of live flammable materials in a forestry environment due to the increasing demand of autonomous solutions in forest maintenance. Specifically, we have implemented modules for semantic segmentation, depth completion and live flammable material localization, and report on the integration of these software components and their performance in operational conditions. Experimental tests show that the integration of all three modules have shown promising results in real time. Semantic segmentation F-1 Score for the ‘fuel’ is near the targeted value and depth completion and fuel localization have great promise with the current limitations. This suggests that our implementation is a step forward in successfully proposing an unmanned ground vehicle (UGV) for forest upkeep.