A Survey on Deep Learning Models for Anomaly Trajectory Detection

The rapid advancement in positioning technology has significantly increased the generation of trajectory data. Unlike tasks such as classification, clustering, prediction, and pattern mining in trajectory analytics, trajectory anomaly detection aims to identify unusual, uncommon, and atypical trajectory behaviors. This detection plays a vital role in various sectors including behavior analysis, autonomous driving, security monitoring, logistics, and freight transportation. This paper offers a comprehensive review of the current state and latest advancements in anomaly trajectory detection techniques within the deep learning domain, highlighting the unique challenges and methodologies in this research area. It begins by analyzing the unique characteristics of anomaly trajectory detection problems as well as the challenges faced by existing research. It then performs a systematic classification and comparative analysis according to existing techniques for detecting anomaly trajectories. The paper also covers commonly used trajectory datasets and evaluation metrics for detection performance assessment. Lastly, the paper summarizes unresolved issues in the field of trajectory anomaly detection and provides insights into future research trends and potential solutions.