Document Type

Data Set

Publication Date

Winter 2024

Abstract

Eco-driving is a driving strategy that focuses on reducing fuel consumption, carbon emissions, and improving passenger comfort through the implementation of safe and anticipatory driving strategies. One method to improve eco-driving in autonomous vehicles is the optimal control approach, which utilizes ‘Signal Phase and Time’ (SPaT) information to determine an energy-efficient trajectory for vehicles approaching and departing signalized intersections. Camera-based vision systems and deep learning techniques can be implemented to recognize the traffic signal head and its phase information, further enabling the vehicle to adapt its driving patterns for improved fuel efficiency and reduced emissions. This report reviews advancements in vision-based traffic signal recognition research, highlighting the advantages of deep learning-based algorithms over traditional computer vision and machine learning techniques, particularly in complex traffic scenarios. An end-to-end deep learning network is developed for traffic light recognition, achieving 94.4% mean Average Precision and an average detection speed of 143 FPS, making it suitable for real-time traffic signal detection in autonomous vehicles. Next, an optimal control approach is explored to devise an optimal speed profile for autonomous vehicles for non-stop crossing approaching signalized intersections. The eco-driving algorithm's effectiveness is evaluated through the SUMO traffic simulation software, showing an 11.11% reduction in vehicle idle-time, 3.85% savings in fuel consumption, and a 3.88% reduction in CO2 emissions with a 100% vehicle penetration rate.

Comments

Tran-SET Project 22ITSLSU41

Download

Share

COinS