A knowledge graph construction method based on co-occurrence for traffic entity prediction

作者：Zhangcai Yin, Yiran Chen, Jiangyan Gu, Shen Ying, Yuan Guo · 发表于：International Journal of Applied Earth Observation and Geoinformation · 年份：2025 · DOI：10.1016/j.jag.2025.104717 · 被引用次数：1 · 研究领域：Traffic Prediction and Management Techniques、Advanced Graph Neural Networks、Graph Theory and Algorithms

• Real-time perception alone fails to meet the safety response requirements of defensive driving. • Co-occurrence relationships in real-time sensing boost predictive awareness of hazardous entities. • Forecasting of risky entities provides critical early warning information and allows a valuable response window. • Co-occurrence knowledge graphs and their probability matrices advance forward-looking perception technology. Co-occurrence can improve the perception and prediction of traffic entities by predicting another traffic entity in automated driving based on a known entity. Existing traffic entity co-occurrence relationship construction methods use a bottom-up approach that relies on labeled datasets. However, the limited nature of datasets constrains the capture of long-tail scenarios. High-definition (HD) maps define the concept of traffic elements and their classifications, and can be used for scene simulation and real-time updating of digital twins. Based on this, this paper proposes a knowledge graph construction method for traffic entity prediction, employing a top-down approach and introducing HD maps. This method establishes a co-occurring semantic network between entities, utilizing prior knowledge. The objective is to reconcile the finiteness of datasets with the infinity of long-tailed scenarios, thereby providing a theoretical basis for the safety calibration of autonomous driving. The experimental results show that the knowledge graph constructed by the top-do...