Author Type

Graduate Student

Date of Award

Spring 4-21-2026

Document Type

Thesis

Publication Status

Version of Record

Submission Date

April 2026

Department

Computer and Electrical Engineering and Computer Science

College Granting Degree

College of Engineering and Computer Science

Department Granting Degree

Electrical Engineering and Computer Science

Degree Name

Master of Science (MS)

Thesis/Dissertation Advisor [Chair]

Evangelos I. Kaisar

Thesis/Dissertation Co-Chair

Zhen Ni

Abstract

To make sure that self-driving and connected automobile technologies are safe and work well, it’s really important that they can correctly identify lanes. But lane detection Algorithms typically have a hard time working well when the weather is bad, such when it rains, fogs, or goes too fast. The circumstances cause visual distortions that make existing computer vision systems less reliable, which makes it harder requires autonomous navigation systems to work well. This paper introduces a comprehensive lane detection system that integrates synthetic Weather-informed data augmentation combined with a Weather-aware Temporal Lane Detection Network (WTLDNet) to make it easier for the model to function in difficult driving conditions. situations. The suggested method starts with the use of fake bad weather adding things like fog, rain, and blur to real driving pictures. The extra instances make the training dataset more diverse and help the model understand qualities that stay the same even when the environment changes. Researchers have created a novel deep learning architecture that combines a convolutional a neural network (CNN) encoder and a transformer-based temporal attention module, which makes it stronger. The CNN backbone uses a number of convovii lutional layers, ReLU activation, and max pooling are used to get spatial lane characteristics from a series of video frames. The transformer module uses multihead Self-attention is used to find the temporal relationships that occur between frames. This mixed design lets the model keep the lanes and get lane structures that might be missed when vision is limited. We use the TuSimple lane detection benchmark dataset for our tests, adding more changes in the weather throughout the training and testing stages. We use a number of different measures to measure how well the model works, such as accuracy and precision. F1-score, recall, and Intersection-over-Union (IoU). A new way to measure performance, Lane Recovery Rate (LRR) has been created to see how well the model works. can restore lane infrastructure even when the weather is bad. The experimental findings demonstrate that the WTLD-Net architecture has been improved. via weather-based augmentation throughout exercise, shows more strength in comparison to baseline models. The model demonstrates its capability to recognize items. better in all kinds of weather and make the rebuilding of lane buildings in tough places. This suggested approach makes the dependability of lane detecting systems, which makes it safer to utilize autonomous and linked automobile technology in normal driving situations.

Download

Share

COinS