Self-supervised Deep Learning Framework for Anomaly Detection in Traffic Data
View recorded webinar on Youtube
The current state of the practice in traffic data quality control (QC) features rule-based data checking and validation processes, where the rules are subjective and insensitive to variation inherited with traffic data. To improve the traffic data QC process, self-supervised deep learning approaches (e.g., VAE, RNN, LSTM, GRU, and LTC) were explored to detect data anomalies manifested in different temporal granularities through embedding. Our approaches leverage the existing multisource traffic data and permits cross-checking of one data source against another for improved robustness. The effectiveness of the proposed methods was demonstrated using two independent sources of traffic data from the Georgia Department of Transportation: (1) traffic counts collected by inductive loops as part of the statewide traffic count program, and (2) traffic volumes acquired by video detection as part of the Georgia 511 system
Jidong Yang, Ph.D., is an Associate Professor in Civil Engineering at the University of Georgia (UGA). He is a registered professional engineer and has over 19 years of combined research and industrial experience in transportation. Dr. Yang’s current research has been heavily focused on AI and deep learning methods and their applications in transportation engineering, especially in ITS, highway design & safety, and sustainable & resilient infrastructure systems. Dr. Yang currently serves on the Editorial Board for ASCE Journal of Infrastructure Systems and the TRB AED50 Committee – Artificial Intelligence and Advanced Computing Applications. He is Review Editor for Frontiers in Future Transportation – Transport Safety and Lead Editor for two special journal editions, focusing on “Machine Learning in Smart Transportation Applications” and “Data-driven Smart Infrastructure Solutions.”
