Citation

Martínez-Vera E, Bañuelos-Sánchez P, Etcheverry G, Vergara-Villegas OO, Cruz-Sánchez VG, Sossa-Azuela JH, Carrasco-Ochoa JA, Martínez-Trinidad JF, Olvera-López JA. Lect. Notes Comput. Sci. 2022; 13264: 25-34.

Copyright

(Copyright © 2022, Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/978-3-031-07750-0_3

PMID

unavailable

Abstract

Autonomous driving systems aim to reduce traffic related accidents; hence, lane changing models need to be developed to accurately represent the driving behaviour during this complex manoeuvre. In this work, extensive pre-processing of real-world detailed vehicle trajectory information from the Next Generation Simulation (NGSIM) dataset was performed in order to derive a suitable feature space to be used as the input data to train and test a lane changing model. State-of-the-art algorithms use the relative velocities and gaps of vehicles in adjacent lanes to derive accurate prediction of the intention to move from one lane to another. Two new feature spaces were developed based on these data and four classification models were implemented by using Logistic Regression, Adaptive Boosting and Extreme Boosting algorithms obtaining accuracies up to 98%.

Language: en

Keywords

Data Modeling; Lane change; NGSIM