Abstract
Aiming to move from conventional throughput-centric paradigms to intelligent, context-aware systems able of perception and autonomous decision-making, sixth-generation (6G) wireless networks is seeking. Driven by recent developments in deep learning and edge artificial intelligence, computer vision (CV) proves to be a key enabler for such perceptive 6G systems. This paper offers a thorough overview bringing together the scattered terrain of CV-enabled 6G technologies. It benchmarks current models against major 6G performance criteria, evaluates architectural paradigms including federated and split learning, and presents a disciplined taxonomy of use cases. This study also notes the possibility of incorporating new technologies with CV to make it more effective, such as fluid antenna system (FAS) and fluid antenna multiple access (FAMA). The study shows that CV integration improves fundamental 6G capabilities like beamforming, mobility prediction, localisation, semantic communication, and immersive control. It also reveals limits in real-time inference under URLLC constraints, data scarcity, and energy economy, though. This work presents a unified basis for advancing CV-native 6G networks by spotting open challenges and suggesting a roadmap including generative perception, collaborative intelligence, and green vision computing.