From Sensors to Vision: Evolving Data Modalities for Intelligent IoT Systems

Abstract

The rapid growth of the Internet of Things has increased the need for sensing systems capable of understanding complex environments beyond simple data collection. The integration of artificial intelligence and computer vision has driven a shift toward perception-oriented Artificial Intelligence of Things (AIoT) systems. This study systematically synthesizes research on vision-based sensing in AIoT, focusing on application domains, sensing modalities, AI methods, and computing architectures. A systematic literature review following PRISMA 2020 guidelines using Scopus identified 24 empirical studies published between 2018 and 2025. The findings show that AIoT vision systems are increasingly applied in manufacturing, agriculture, infrastructure, and surveillance, supporting real-time monitoring and decision-making. Core functions include detection, classification, segmentation, and activity recognition, enabled by deep learning and edge–cloud architectures. The results indicate a shift toward multimodal sensing and edge intelligence, highlighting a broader transition to perception-centric AIoT systems, with ongoing challenges in dataset generalization, multimodal integration, and efficient edge deployment.