Abstract: To address the phenomenon of limited pedestrian vision caused by smoke and insufficient lighting during emergencies, as well as the shortcomings of traditional evacuation models such as omniscient perception and simplified guidance effects, this paper proposes and implements a crowd evacuation model incorporating limited vision and guide intervention. The model constructs a clear hierarchical "perception–decision–movement" architecture by integrating the precise motion description capabilities of the social force model with the autonomous decision-making abilities of Agent theory. Building on this architecture, the model introduces a pedestrian visual field limitation mechanism that confines Agent perception to a specific range, thereby more realistically simulating the constrained cognition of pedestrians in emergencies. Furthermore, it designs and implements a dual-mode (static and dynamic) guide intervention mechanism, establishing a refined interaction model between guides and ordinary pedestrians. Experiments analyze the effects of visual radius variation, guide compensation, and guide initial positioning on evacuation efficiency. Simulation results demonstrate that: Visual radius significantly impacts evacuation efficiency—in the test scenario, increasing the radius from 2m to 4m reduced evacuation time by 42.8%; Guides effectively compensate for limited vision, with static guides proving more effective than dynamic guides under severely restricted vision (2-3m), reducing evacuation time by approximately 4.34 seconds and 2.45 seconds, respectively; The initial position of guides plays a critical role in improving efficiency—static guides achieve optimal performance at the room center, while dynamic guides exhibit optimal effectiveness when positioned on the side away from the exit.