Research on trajectory compression algorithm based on deep learning

In recent years, with the development of intelligent transportation network and the improvement of people's travel requirements, in order to realize the intelligent operation mode of taxi, it is essential to analyze the vehicle trajectory. In order to analyze the trajectory of the aircraft more...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2024 5th International Conference on Computer Vision, Image and Deep Learning (CVIDL) s. 625 - 630
Hlavní autoři: Zhao, Xinhui, Yang, Liu, Liu, Heng
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 19.04.2024
Témata:
Computational modeling
Data models
Deep learning
Informer
Predictive models
StemGNN
Time series analysis
Trajectory
Trajectory compression
Transformers
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:In recent years, with the development of intelligent transportation network and the improvement of people's travel requirements, in order to realize the intelligent operation mode of taxi, it is essential to analyze the vehicle trajectory. In order to analyze the trajectory of the aircraft more effectively, the trajectory compression technology is added to extract the important features of the trajectory and reduce the amount of data. By analyzing the spatiotemporal characteristics of vehicle GPS track data, a lossless compression model based on prediction is adopted in the process of track compression. In this paper, a trajectory multi-step prediction model (SG-Informer) based on deep learning is proposed. The model transfers the time domain to the frequency domain by using a spectral time graph neural network (StemGNN) and captures the time and space dependencies in the frequency domain. Gated cycle unit (GRU) is used to solve the dependence of the captured time distance in the time series, and then the calculation results are passed to Informer, which makes multi-step recursive prediction of the time series, and finally sets the compression ratio and compresses according to the size of the error. The model is verified by using Chengdu taxi GPS data, and the results show that the prediction effect of the model is better than that of LSTM, Transformer and Informer models.
DOI:10.1109/CVIDL62147.2024.10603753