Abstract: Aiming at real-time and accurate pupil detection, a master-slave camera tracking eye image capture system was proposed to capture eye images. The master camera was arranged to locate the face, and the slave camera was arranged to track the eye area in real time under the cloud platform and to collect high-resolution eye images. In order to improve the dynamic performance of the system, a fusion algorithm of AdaBoost face detection and MOSSE target tracking was adopted for the face image with the resolution of 1280×960 from the master camera, realizing a fast face location of 60 fps. A semantic segmentation network was used to locate the pupil diameter of the eye image from the camera to obtain the region of interest of the pupil. The pupil of the interest region was detected accurately with an ellipse fitting algorithm. The experimental results show that the proposed detection system and method can achieve 18 fps pupil detection. The effective pixel of pupil image can reach more than 2 500, and the detection repeatability can reach ±1.06%.