A Cooperative Vehicle-Road System for Anomaly Detection on Vehicle Tracks With Augmented Intelligence of Things
Yuxin Zhang, Limei Lin, Yanze Huang, Xiaoding Wang, Sun‐Yuan Hsieh, Thippa Reddy Gadekallu, Md. Jalil Piran
Abstract
The Augmented Intelligence of Things (AIoT) is an emerging technology that combines augmented intelligence with the Internet of Things (IoT) to facilitate advanced decision-making processes. In this paper, we focus on the detection of vehicle trajectory anomalies in a vehicle-road collaboration system by AIoT, aiming to improve the traffic safety and road operation efficiency. We transmit collaboration data collected by sensors to an IoT server, which enables the effective data analysis for vehicle trajectory information. We propose a self-supervised learning augmented intelligence algorithm to achieve precise and efficient detection of trajectory anomalies. First, we models the traffic road network as a topology graph. Subsequently, we sample the relevant subgraph contexts for each target node through a random walk algorithm. And the subgraphs with higher intimacy scores are selected as the contextual background to be input along with the target node. After that, the anomaly score of each target node is computed through the generative learning module and the contrastive learning module. To evaluate the effectiveness of our anomaly detection approach, we initially conduct pre-training of the model using four widely utilized graph machine learning datasets. The experimental results reveal that our approach surpasses previous methods in the accuracy of identifying graph anomaly nodes. In addition, we carry out our approach on two real traffic datasets with high accuracies of 86.47% and 85.2%, respectively. This result demonstrates the effectiveness of our proposed approach in detecting trajectory anomalies in real traffic scenarios.