Aiming at the problem that people with different diopters cannot freely change lenses when using wearable near-eye eye-tracking devices, this paper developed a set of plug-in near-eye eye-tracking system equipment. The system is composed of two parts: a near-eye acquisition hardware platform and an eye movement feature extraction algorithm. The hardware platform is an image acquisition module designed according to the structure of the optometry trial frame. The eye movement feature extraction algorithm is to train the eye detector and eye feature point detector based on machine learning, and then combines the candidate pupil fitting screening strategy to obtain the pupil diameter, center point, and blink frequency. The experimental results show that in this system algorithm, the correct rate of pupil detection is 97.24%, and the correct rate of blink detection is 91.59%. The device can meet the needs of eye-tracking detection and research for people with ametropia, and provide accurate and reliable eye-movement data.