Commonsense reasoning in computer vision refers to the integration of visual data with contextual knowledge to improve a model's grasp of everyday scenes. This goes beyond conventional CNNs that merely detect objects, enabling holistic scene interpretation, spatial relationship inference, and action understanding, which leads to more accurate predictions and interactions in real-world settings.\ \ The survey categorizes the main approaches for injecting commonsense into vision tasks. Knowledge‑graph methods supply structured entity‑relation information; scene‑graph techniques build intra‑image graphs of objects and their interactions; neuro‑symbolic models combine neural perception with symbolic logical reasoning; and commonsense‑augmented Transformers incorporate external knowledge vectors into self‑attention mechanisms to boost cross‑modal reasoning.\ \ Key challenges remain, including dataset bias, incomplete knowledge bases, and difficulties in seamless multimodal integration. Prospective directions point to cross‑modal reasoning frameworks, scalable commonsense injection pipelines, and deeper neuro‑symbolic hybrid architectures, all aimed at creating truly intelligent visual systems.\ \ Review