To address the limitations of traditional image-based visual servoing (IBVS), such as its heavy reliance on the image Jacobian model, insufficient adaptability under fixed gains, and susceptibility to feature-point loss from the camera field of view, this paper proposes an improved fuzzy adaptive online sequential extreme learning machine image-based visual servoing method (FA-OS-ELM-IBVS). The proposed method employs an online sequential extreme learning machine (OS-ELM) to directly estimate camera velocities from image errors, thereby circumventing the explicit computation of the image Jacobian and its associated singularities. A Mamdani-type fuzzy gain regulator, which takes the error norm, manipulability, and error convergence angle as inputs, is designed to enable nonlinear and adaptive adjustment of servo gains. Moreover, a hierarchical rectangular region strategy combined with sigmoid-based smooth compensation is introduced to achieve continuous and controllable field-of-view maintenance. The stability of the proposed control system is rigorously analyzed and proven based on the Lyapunov theory. Simulation results demonstrate that, compared with the conventional IBVS, the proposed method reduces the convergence time by approximately 7.09%–16.7% and shortens the camera trajectory length by about 27.48%–57.94%, while substantially reducing integral performance indices such as IAE and ITAE. Notably, stable convergence is maintained even under a pronounced mismatch between the assumed and actual depths. Further comparisons with representative improved IBVS schemes confirm superior performance in both convergence speed and servoing accuracy. Experiments conducted on a six-DoF CGXi G6 robot platform corroborate these findings, verifying the improved convergence efficiency in real-world scenarios and demonstrating the robustness and effectiveness of the proposed uncalibrated framework.