Visual Anomaly Detection for Industrial Environments with Three-Dimensional Structural Complexity

The aim of this thesis is to explore visual anomaly detection solutions for industrial environments, with particular emphasis on quality control scenarios characterized by a strong intrinsic 3D nature. Realistic systems are affected by challenges arising from the unconstrained degrees of freedom in space, which demand robust and reliable methods capable of handling complex 3D variability in real-world industrial settings. The thesis is structured as follows: it first reviews key concepts in artificial intelligence and computer vision, which provide the background for the proposed visual quality control systems. Then, two complementary scenarios based on realistic industrial settings are presented, describing the proposed architectures, providing qualitative and quantitative results, and comparing them with other state-of-the-art methods.