Spoluřešitelé:
Anotace:
This research project aims to address the challenges in developing deep learning-based computer vision applications for intelligent autonomous vehicles (IAVs) under adverse weather conditions, particularly in the presence of haze. Existing datasets used to train IAVs have limitations in capturing the diversity of haze conditions, leading to reduced performance in such weather. To overcome these limitations, the project introduces a novel haze dataset generation method called the Domain Flow Adaptation Network (DFA-Net). DFA-Net leverages semantic information from clear images and haze representations from reference hazy images to generate high-quality hazy images with diverse density levels. The project prioritizes data integrity and security for both real-world and synthetic data in the development of computer vision applications for IAVs, with a focus on blockchain technology to ensure data authenticity and prevent breaches. These augmented datasets can significantly improve computer vision applications such as object detection and target tracking in hazy conditions. The project includes various modules such as Semantic Extraction, Haze Extraction, Image Production, and Image Assessment. Real-world experiments will be conducted to evaluate the effectiveness and applicability of the proposed approach. Additionally, this is a bilateral project between the Czech Republic and Taiwan, aimed at fostering technology exchanges and contributing to the advancement of knowledge in the field. The project further plans to disseminate its findings by publishing in top journals, ultimately aspiring to apply the research to benefit industrial products and real-world applications.
Pracoviště:
Rok:
2024 - 2024
Program:
Společné výzkumné projekty mezi TAIPEI TECH a ČVUT
