Towards Minimizing Domain Gap When Using Synthetic Data in Automotive Vision Control Applications
Mohamed Slim Werda, Hamza Taibi, Khalid Kouiss, Ahmed Chebak, Saif Ben Halima, Michael Decottignies, Carey Dilliott · IFAC-PapersOnLine · 2024
This paper explores the generation and use of synthetic data in the development of AI-based vision control systems within the automotive industry. Addressing the challenges of collecting diverse and high-quality datasets, we demonstrate the effectiveness of synthetic data in training deep learning models for defect detection and quality inspection. By leveraging controlled randomizations during image dataset generation, we mitigate the domain gap between synthetic and real-world data, enhancing model performance and generalization. Our approach, validated across multiple industrial applications, significantly accelerates the development cycle and improves the accuracy of AI-based inspection systems. This study highlights the potential of synthetic data to overcome data scarcity and improve the efficiency and effectiveness of AI-driven quality control in automotive manufacturing.