Since most of today's eye trackers are based on infrared reflections which means the eye tracker needs a camera, it also needs an extra infrared light gadget. This paper refers to the shortcomings of modern infrared eye-tracking devices. And then design an experiment to collect eye-tracking datasets using only a camera and implement an eye-tracking system deployed on a head-mounted human-computer interaction device by using image recognition techniques and key-point detection techniques. This method makes the eye tracker more complicated. This paper designs an experiment by getting the eyeball data from different people in different environment backgrounds to create a new eye-tracking dataset for head-mounted interactive devices and transform that dataset into a '.txt' format file usable by YOLO. Then use this dataset to finish the transfer learning which is based on the YOLOv8, which trained a theoretically realisable model of an eye-tracker. Thereby, the efficiency of the usage of the eye-tracker is optimized in principle, allowing people to use the eye-tracker without relying on an infrared device, but only a camera.