Proactive Congestion Detection in V2V Scenarios With an Integrated Multi‐Layer Information Fusion Approach

ABSTRACT

Real‐time traffic congestion detection is critical for supporting effective traffic management and control decisions. To improve traffic congestion detection accuracy in V2V scenario with limited penetration rate (PR) of connected vehicles (CVs), the present work proposes a proactive traffic congestion detection scheme based on a multi‐layer information fusion framework. Specifically, in a V2V scenario with both CVs and non‐CVs, each CV has the ability to estimate the local congestion condition through the feature‐level information fusion based on machine learning algorithms, using the motion information of itself and its surrounding CVs within the communication range. Then the local congestion estimation results of each CV in a certain region are applied to estimate regional traffic congestion state through decision‐level information fusion based on DS evidence theory. The feasibility and accuracy of the proposed approach under different PR settings of CVs are then validated using the HighD dataset. The results show that the proposed multi‐layer information fusion method achieves great performance even under low CV PR situations, and the congestion detection accuracy could reach 92.06% when the CV PR is only 20%. Moreover, the comparative experiment results indicate the proposed approach outperforms other two baseline models, which highlights the significance of mining the underlying relationship between CVs' microscopic driving behaviour characteristics and fine‐grained road traffic states in proactive congestion detection. The outcomes of the present work can provide valuable insights for the further application and improvement of real‐time traffic estimation in V2V scenarios.