Automotive Safety: a Neural Network Approach for Lane Departure Detection using Real World Driving Data
Paper in proceeding, 2019

Lane departures, where the vehicle leaves the lane due to driver inattention, drowsiness, or incorrect situation assessment, are one of the most serious accident and fatality prone scenarios. To further improve traffic safety, we are asking the question: How much can a neural network approach improve the reliability of lane departure predictions compared to traditional model-based methods? Our results show a relative improvement in reliability of 7% in terms of true positive rate and 22% reduction of the false positive rate with respect to a constant velocity model method. The key contributions of this work are the introduction of sparse sampling in the input data, a thorough comparison with a baseline solution, and the evaluation on real world driving data.


John Dahl

Zenuity AB

Rasmus Jonsson

Zenuity AB

Anton Kollmats

Zenuity AB

Gabriel Rodrigues de Campos

Zenuity AB

Jonas Fredriksson

Chalmers, Electrical Engineering, Systems and control

IEEE Conference on Intelligent Transportation Systems, Proceedings, ITSC

2019 IEEE Intelligent Transportation Systems Conference (ITSC)
Auckland, New Zealand,

Areas of Advance


Subject Categories

Transport Systems and Logistics

Infrastructure Engineering

Vehicle Engineering



More information

Latest update

6/2/2020 6