Abstract

As the demand for self-driving cars grows, the reliability of traffic sign recognition is essential for commuter safety. Researchers have explored several machine-learning and deep-learning ap- proaches to traffic sign identification, but Vietnam's unique traffic environments, ranging from com- plex urban intersections to highways with vertically stacked signs, present unique challenges. While conventional object detection techniques can handle typical urban traffic signs, they struggle with the groups of stacked signs that are commonly found on Vietnamese highways. This study ad- dressed this problem by treating each detected sign as a node in a graph and modeling its spatial and semantic relationships with edges using Graph Neural Networks, which can learn to identify patterns and groupings. This approach not only allows for the accurate detection of each sign but also captures the collective intent of grouped signs in both urban and highway contexts, thereby providing commuters with more reliable and contextually aware guidance when navigating Viet- nam's complex traffic sign system.