Research on vehicle trajectory anomaly detection algorithm based on GRU and WGAN
Yuhang Liu, Lei Wang, Xiaoyong Zhao, HuaMing Lu, JingLe Zhang, Jianhua Li, DeBin Han · 2023
The uncertainty of vehicle trajectories and the existence of anomalous data lead to challenges in their application in the field of digital transportation. In this paper, GRU-WGAN deep learning model based on GAN is proposed for vehicle trajectory feature extraction and anomalous trajectory detection. Firstly, VAE utilises the GRU neural network as the Encoder and Decoder part, which can deeply extract features from the original data at the encoding layer and do variational inference. At the same time, learning deep feature extraction helps the VAE model to restore the approximate probability distribution of the initial data at the coding layer to the maximum extent, thus improving the efficiency of anomaly detection. The GRU-WGAN model combining GRU and WGAN is then proposed to learn the output of the feature extraction part and the potential features of the real data to complete the task of anomaly detection of vehicle track data. In addition, comparative experiments were set up to validate the proposed model. The experiments demonstrate that the GRU-WGAN model outperforms the conventional algorithm in terms of accuracy, recall and F1 metrics. Therefore, the proposed model can be effectively applied to feature extraction and vehicle trajectory anomaly detection tasks.