Vehicle Trajectory-Based Intersection Classification Using Deep Transfer Learning

Abstract

This document is a model and instructions for M-TEX. The accelerated expansion of global road networks necessitates the availability of accurate and up-to-date cartographic data for connected and automated vehicles. Conventional mapping techniques, such as satellite imagery and field surveys, are inher-ently time-consuming and labor-intensive. This study proposes a novel approach that makes use of the ubiquity of GPS-embedded devices to facilitate the efficient updating of road maps. We present a classifier that identifies the types of intersections from vehicle trajectories, which represents a crucial initial step in the generation of automated maps. The methodology employs five distinct convolutional deep neural network architectures for the detection and classification of road intersection types. The experimental results, which were obtained using three real-world vehicle trajectory datasets, demonstrate classification accuracy levels that range from 84 % to 91 %. It is noteworthy that these results were achieved using less than 10% of the data that is typically required for comparable accuracy levels, which highlights the efficiency of the feature extraction approach that was employed. This research contributes to the advancement of intelligent transportation systems by offering a more efficient and scalable method for road network mapping and updating.