Modeling Trajectories Obtained from External Sensors for Location Prediction via NLP Approaches

Area of Science:

• Artificial Intelligence
• Machine Learning
• Transportation Systems

Background:

• Representation learning extracts low-dimensional features from high-dimensional data.
• Natural Language Processing (NLP) models, like word embeddings and language models, capture semantic meaning and sequential order in text.
• Trajectory data in transportation networks presents a complex, high-dimensional challenge for representation learning.

Purpose of the Study:

• To adapt NLP techniques for trajectory representation learning.
• To encode road network sensor data using NLP methods.
• To predict the next vehicle movement based on learned trajectory features.

Main Methods:

• Utilized NLP methods inspired by word embeddings and language models.
• Encoded external sensor data from the road network.
• Generated a feature space for trajectory representation.
• Evaluated vector representations using extrinsic and intrinsic strategies.

Main Results:

• Demonstrated the effectiveness of NLP models in encoding road sensor data.
• Successfully generated a feature space for trajectory analysis.
• Achieved accurate next vehicle location prediction.
• Validated the potential of NLP for trajectory applications.

Conclusions:

• Natural Language Processing models show significant potential for trajectory representation learning.
• The proposed method enables effective prediction of future vehicle movements using sensor data.
• This approach offers a novel way to analyze and understand complex transportation dynamics.