Toward automated traffic sign condition evaluation utilizing vehicle-mounted cameras and LIDAR
No Thumbnail Available
Authors
Meeting name
Sponsors
Date
Journal Title
Format
Thesis
Subject
Abstract
This study proposes a methodology to automate and streamline traffic sign condition assessments and introduces a numerical rating system to evaluate the visual and structural integrity of signs in both rural and urban environments. The goal is to support transportation agencies in identifying damaged, obstructed, or deteriorating traffic signs that may reduce visibility or driver comprehension. The methodology develops a Sign Condition Index (SCI) based on key features, enabling consistent and versatile evaluations. Traffic sign data was collected using an Insta360 ONE X2 camera mounted on a vehicle to capture a video of roadways in and around Columbia, Missouri. Extracted frames were annotated using CVAT (Computer Vision Annotation Tool) to label traffic sign types, and these cropped images were analyzed by a pre-trained large language model (LLM), which generated concise descriptions across four key categories: legibility, color, shape integrity, and surrounding environment. Following manual corrections of the LLM-generated responses, the pairs of cropped images and corrected responses were used to fine-tune the original LLM. This trained version of the model was better equipped to recognize and describe subtle visual features such as fading or physical damage, improving the reliability of automated assessments. In a second data collection phase, a GoPro Hero 11 and a Livox HAP LIDAR sensor were mounted together to capture high-resolution video and dense 3D spatial data. The LIDAR and GoPro datasets were fused and used in model training, enabling automatic detection and classification of signs. Following detection, each sign's condition was scored using the Sign Condition Index. Signs were assigned a value out of four based on yes/no evaluations of the LLM-generated attributes. The process provides a repeatable framework for large-scale traffic sign assessment and maintenance prioritization. A case study utilizing the methods is provided.
Table of Contents
PubMed ID
Degree
M.S.