As a core sensing and decision-making technology in industrial and automation domains, machine vision is required to maintain high-precision image-processing performance while simultaneously meeting multiple engineering constraints, such as real-time operation, miniaturization, and energy efficiency. These demands have driven the deep integration of machine vision and edge intelligence, giving rise to a new research frontier—real-time intelligent machine vision systems. This article systematically reviews the progress of this interdisciplinary field from three perspectives: algorithm evolution, hardware-software co-design, and engineering applications. First, it summarizes the development of intelligent vision algorithms represented by convolutional neural networks and vision transformers, emphasizing the role of model compression techniques—including pruning, quantization, and lightweight architecture design—in enabling edge deployment. Second, it explores edge computing platforms suitable for intelligent machine vision tasks and discusses typical strategies for hardware-software co-optimization in deploying artificial intelligence algorithms. Furthermore, through analyses of representative applications such as intelligent connected vehicles, urban surveillance systems, and robotics, the paper highlights the advantages and potential of edge vision systems in latency-sensitive scenarios. Finally, it identifies current challenges, including system integration, computational efficiency, energy efficiency, and intelligence level, and envisions emerging directions such as in-sensor computing, in-memory computing, neuromorphic computing, and on-device deployment of large models at the edge. Overall, this review emphasizes that the deep synergy among algorithms, hardware, and applications will be the key pathway toward realizing autonomous, adaptive, and perceptually intelligent machine vision empowered by edge intelligence.