Manual implementation of retinal microsurgery is extremely difficult and poses strict requirements for doctors’ operations. The clarity and flexibility of intraoperative images play an important auxiliary role in the operation. Microscopes and Optical Coherence Tomography (OCT) devices can provide image information from a fixed perspective. However, when the patient’s cornea is cloudy or the pupil atrophies, clear intraoperative images cannot be obtained. Endoscopes can be inserted into the eyeball through scleral wounds to avoid the problem. However, the handheld rigid ophthalmic endoscope is difficult to obtain intraoperative operation images from multiple perspectives within the eye and also adds an extra burden for doctors to perform intraocular surgical operations. This paper proposes an intraocular continuum endoscope actuator (ICEA) with two degrees of freedom (2-DoF). It can provide multi-view intraoperative images during movements and observe the posture of surgical instruments. Furthermore, this paper proposes a visual control method based on endoscopic images to enable to follow the end of surgical instruments during the operation. The YOLOv8n object detection algorithm is used to obtain the feature coordinates of the microneedle ends in the endoscopic image plane. The actuator and the following control method have been verified through eye model experiments on the eye surgical robot (ESR). The experimental results prove that it can stably track the end of the instrument, making it converge to the expected coordinate.