Classification of Lane changing behavior from trajectory data using self-supervised learning method

This project seeks to develop a system to accurately identify car-following (CF) and lane-changing (LC) behaviors using trajectory data from the pNEUMA dataset, an urban trajectory dataset collected from drones. Previous methods relied on fixed thresholding or manual labeling to separate the phases of a vehicle’s trajectory (CF or LC), which can be time-consuming and less accurate. We aim to develop a generalizable and accurate model that automatically identifies different phases of car-following and lane-changing behavior without requiring manual labeling. To achieve our goal, we will employ a self-supervised learning method that involves training an auto-encoder on the dataset to capture underlying patterns and features without manual labeling to classify different behaviors. The system will be evaluated on the pNEUMA dataset, expecting high accuracy in behavior identification. We expect the results to demonstrate the effectiveness of self-supervised learning in classifying complex driving behaviors. The proposed method can be further used in transportation planning, driver assistance systems, and autonomous vehicles in urban environments.

Faculty Supervisor:

Lijun Sun

Student:

Partner:

École polytechnique fédérale de Lausanne

Discipline:

Engineering

Sector:

Education

University:

McGill University

Program: