Education
- School of Transportation, Southeast University, Phd, 2019.06-2025.03 (expected)
- School of Automation, Nanjing University of Science and Technology, Bachelor's, 2015.09-2019.06
Research Interest
- Intelligent Transportation Systems, Computer Vision, Vehicle Detection, Accident Detection/Prediction, Multi-modal Large Models
- vvgod@seu.edu.cn
Bio
Wei Zhou is a Ph.D. candidate at Southeast University, advised by Prof. [Chen Wang]((https://trmetagroup.github.io/team/ChenWang). He is currently in the fourth year of his doctoral studies, expected to graduate before March 2025.
As of now, he has published/accepted a total of 25 academic papers, including both journal and conference papers. It is important to note that he has published 11 papers as the first author/corresponding author in top journals, such as IEEE Transactions on Intelligent Transportation Systems and Automation in Construction, with 9 of these papers in JCR Q1 journals, including 5 papers published in IEEE TITS as the first author.
Moreover, he also serve as a young editorial board member for Digital Transportation and Safety and am a reviewer for top journals including IEEE TITS, IEEE Transactions on Intelligent Vehicle (IEEE TIV), IEEE Transactions on Vehicular Technology (TVT), Accident Analysis and Prevention (AAP), Advanced Engineering Informatics (AEI), Engineering Applications of Artificial Intelligence (EAAI), Journal of Cleaner Production (JCP), and Neural Networks (NN).
His research interest mainly includes the application of computer vision technology in the transportation field:
- 2D/3D vehicle detection and tracking: Focuses on visual sensors and LiDAR for roadside monitoring, targeting deficiencies of existing detection/tracking methods under adverse weather/lighting conditions and complex traffic scenarios; Utilizes GAN generative methods and small object detection techniques to improve existing methods.
- Accident (road anomaly) detection/prediction: Implements rapid accident detection by constructing (lightweight) deep networks to capture accident appearance features (such as vehicle damage) and motion features (sudden decrease in speed); Achieves accident prediction through group relationship modeling and spatiotemporal reasoning.
- Few-shot learning & Domain adaptation: Addresses the dependence of deep learning models on large data sets and the difficulty of data collection in some scenarios by researching few-shot learning and domain adaptation methods that require only a small number of samples to train and converge the model.
- Pedestrian crossing intention prediction&Trajectory prediction: Achieves pedestrian crossing intention recognition and trajectory prediction by modeling the relationships of the group around the pedestrian and the pedestrian’s own attributes (such as posture, appearance, and historical trajectory).
- Multi-modal large models: Large models, also known as foundation models, possess a wealth of prior knowledge after being trained on vast amounts of data, demonstrating powerful zero-shot and few-shot capabilities. Their potential in the field of traffic perception awaits further exploration.
