Semantic Mapping of Large-Scale Outdoor Scenes for Autonomous Off-Road Driving

Abstract

Semantic mapping is a very active and growing research area, with important applications in indoor and outdoor robotic applications. However, most of the research on semantic mapping has focused on indoor mapping and there is a need for developing semantic mapping methodologies for large-scale outdoor scenarios. In this work, a novel semantic mapping methodology for large-scale outdoor scenes in autonomous off-road driving applications is proposed. The semantic map representation consists of a large-scale topological map built using semantic image information. Thus, the proposed representation aims to solve the large-scale outdoors semantic mapping problem by using a graph based topological map, where relevant information for autonomous driving is added using semantic information from the image description. As a proof of concept, the proposed methodology is applied to the semantic map building of a real outdoor scenario.