Masterarbeiten (abgeschlossen)
-
Traffic Participant Behavior Prediction based on Dynamic Graph Neural NetworkVehicle trajectory prediction is crucial for intelligent transportation systems (ITS) to improve traffic efficiency, alleviate congestion, and enhance safety. Despite advancements in deep learning, trajectory prediction faces challenges due to dynamic agent interactions and complex traffic environments. Traditional rasterized approaches suffer from high computational costs and information loss, prompting a shift towards vectorized methods that use Graph Neural Networks (GNNs) for better feature learning and robustness. We propose a lightweight model using dynamic graph neural networks (DGNN) to improve efficiency, interpretability, and accuracy. Evaluated on the Argoverse 2 dataset, our approach demonstrates reduced computational time, lower parameter count, and high prediction accuracy compared to baseline models. Additionally, we explore a Detection Transformer (DETR)-basedLeitung: XUTeam:Jahr: 2024
Offene Masterarbeiten
-
Dynamic Multi-Agent Motion Prediction with Redundancy Reduction and History-Aware InteractionLearning robust representations of the driving environment is a highly challenging task in autonomous driving systems. Current multi-agent motion prediction methods heavily rely on deep neural networks to process rich spatio-temporal traffic information. Transformer-based and graph attention network-based approaches have been widely adopted to model the interactions among traffic participants and between participants and static map elements, followed by trajectory decoders for future behavior prediction. However, as the performance of deep learning models often scales with the amount of training data, the collection and annotation of large-scale datasets can be prohibitively expensive. Consequently, self-supervised learning methods have gained increasing attention, as they can generate supervision signals from unlabeled data.Leitung: Yiming XuJahr: 2025
-
Detecting anomalous sensor data and vehicle trajectories using deep learning methodsAnomaly detection in vehicle trajectories is a critical task in autonomous driving and intelligent transportation systems. Existing methods primarily rely on rule-based approaches or shallow statistical models, which struggle to capture complex spatio-temporal patterns in modern traffic environments. Recent advances in deep learning have demonstrated strong potential in motion prediction, but their application to anomaly detection in vehicle trajectories remains underexplored. This research aims to develop a deep learning-based trajectory anomaly detection framework, focusing on two types of anomalies. The first type arises from sensor errors, leading to abnormal trajectory data that do not reflect the actual vehicle behavior. The second type concerns driving anomalies caused by human errors, such as traffic violations or accidents. The proposed approach will leverage a motion prediction model as a foundation: when the predicted trajectory significantly deviates from the observed trajectory, an anomaly is reported. Furthermore, given the increasing demand for data privacy and the diversity of data sources across different regions and platforms, this research will incorporate a federated learning framework. Federated learning allows collaborative model training across multiple datasets from different organizations or vehicle fleets without sharing raw data. This ensures data privacy while enabling the detection model to generalize to various environments and sensor configurations.Leitung: Yiming XuJahr: 2025
