Avaluació de la robustesa de sistemes de reconeixement de senyals de trànsit davant degradacions d’imatge
Year: 2026
Type: dissertation
Abstract: This final degree project evaluates the performance of a Traffic Sign Recognition (TSR) system under different image degradations in order to simulate various real world driving scenarios. Currently, many modern vehicles incorporate Advanced Driver Assistance Systems ADAS. These systems assist the driver during the driving process and contribute to improve road safety, from increasing driving comfort to helping prevent potential accidents. Traffic Sign Recognition systems are a key technology within ADAS, as they allow the identification and interpretation of traffic signs from images captured by cameras installed in the vehicle. Although these systems can achieve a good performance under ideal conditions, their accuracy can be affected by different factors that degrade the quality of the captured image, such as low illumination, noise, low resolution or image blur. The principal objective of this project is to analyze the robustness of a traffic sign recognition system under different image degradations that simulate real driving conditions. To achieve this, a public dataset of traffic signs and an AI classification model design implemented in Python are used. Based on this system, several experiments are conducted by applying controlled degradations to the images to measure the performance of the model using accuracy indicators. The result obtained of each simulation allows the study of how different conditions may affect the robustness of the system in correctly recognizing traffic signs, and provide insights into its reliability in real-world environments. Additionally, a simulation of a real driving route is proposed in order to evaluate the system performance under conditions closer to real driving. The simulation is based on a predefined route, where different scenarios and types of roads are reproduced, allowing the analysis of the system in a more realistic dynamic context.
Author: Sun Qiu, Junjie
Language: Catalan
Subfield: Automotive Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7164619708