Co-occurrence knowledge graph construction based on corner cases

In complex urban roads, it is difficult for automatic driving systems to accurately perceive dynamic scene information such as traffic participants facing corner cases due to such problems as obstacle occlusion and complex traffic participants. This makes it difficult for autonomous driving systems to accurately assess the source and level of risk, which seriously affects the safety of autonomous driving. Co-occurrence can link the associated entities in the scene to provide additional semantic information for autonomous driving, and provide corresponding over-the-horizon perception for effective decision-making and safe driving. Knowledge graph describes knowledge in a structured way using the form of entities and edges that can explicitly represent traffic participants, roads, and the relationships between them. In this paper, we propose a cooccurrence knowledge graph construction method based on edge cases according to the co-occurrence relationships between entities in a traffic scene. This method (1) introduces high-definition map ontologies to construct a knowledge graph that describes the directional co-occurrence relationships between entities in the corner cases, and (2) combines the co-occurrence knowledge graph to predict the entities that may appear in the traffic scenes. The experimental results show that the co-occurrence knowledge graph based on the corner cases can improve the automatic driving systems’ understanding of the corner cases and predict the entities that may exist in the scene, providing effective safety assurance for autonomous driving.