This paper proposes a novel 2-DOF stabilization platform for vision application of robotic fish to enhance the stability of image and reject the periodic disturbance from yaw and roll channels caused by fish’s swimming. The problem formulation and system framework of camera stabilization are first discussed. In order to achieve better control effect, forward and inverse kinematics of the 2-DOF gimbal are derived, which combine the feedback of IMU and target states to calculate controllers’ ideal input. Meanwhile, linear active disturbance rejection control (ADRC) without tracking differentiator is adopted in our system, on account of its superior performance to compensate uncertainties and disturbance. Finally, experimental results demonstrate that the error angle of ADRC is obviously smaller than PD and feedback-feedforward control. Furthermore, compared with 1-DOF stabilization platform, the 2-DOF one exhibits the overwhelming advantage about the enhancement of image stability.