Abstract: In constrained environments, image-based visual servoing is prone to feature boundary violations, abrupt image-trajectory changes, and discontinuous execution velocities under large initial deviations and near field-of-view boundaries. To address these issues, this paper proposes a dual-layer visual servoing method that coordinates a virtual feature protection layer with a predictive optimization layer. First, a virtual-feature protection mapping is constructed in the protection layer. When feature points enter a predefined danger region, virtual-feature feedback is used to guide the control input, gradually pulling the image features back into the safe FOV. Second, a hard-constrained quadratic programming controller based on model predictive control is designed in the optimization layer, where FOV constraints, image-step constraints, and velocity change-rate constraints are uniformly incorporated into the optimization process. Experimental results demonstrate that the proposed method effectively reduces FOV violations and suppresses abrupt control-input variations while maintaining feature visibility, thereby significantly improving the continuity of motion trajectories in both image space and Cartesian space.