This article introduces a velocity observer-based adaptive visual servoing control scheme for closed-architecture robotic systems. A novel dual-observer framework, consisting of an image-space observer and a joint-space sliding observer, is developed to avoid the usage of both image-space velocity and joint velocity. Leveraging this framework, an adaptive visual servoing loop control solution is designed, which dynamically generates velocities estimations-based joint position or velocity commands for the torque control loop. Rigorous Lyapunov-based stability analysis demonstrates the asymptotic convergence of both image-space tracking error and observation error, without necessitating modifications to the inner torque controller. This feature renders the proposed method particularly suitable for closed-architecture robotic systems, where the torque controller is unmodifiable due to safety/commercial strategy, and restricted to accepting only joint position or velocity commands. Experimental validation on a robotic manipulator demonstrates the superior performance of the proposed approach.