Abstract: The gaze tracking technology is widely used in many fields, and it has a broad application prospect in the field of human-computer interaction. The technology is based on the eye characteristic parameters and the gaze parameters, and it estimates the direction of sight and placement of sight based on the eye model. Therefore, accurately locating the pupil position is important in the gaze tracking technology, and it directly affects the accuracy of the gaze tracking result. Presently, there are numerous algorithms used in eye detection; however, most of them are characterized by some problems, such as the low accuracy of locating the pupil position, high detection error, and slow operation speed; thus, they cannot meet the accuracy requirements of locating the pupil position. To solve these problems, in this study, a concept of pupil localization method from rough to precise was adopted, and a high-accuracy pupil localization method based on image processing was proposed. In this method, first, the improved maximal between-cluster variance algorithm used the histogram of the pupil region to adaptively segment region to roughly locate the pupil region. Then the pupil edge points can be accurately located by the gradient of the pupil grayscale. Finally, a sub-pixel localization method was adopted on the basis of the pixel level edge points of the pupil to locate the sub-pixel level edge points of pupil more accurately, and the center position of the pupil was accurately determined by the method of ellipse fitting. In addition, an equidistance pupil compensation method was proposed in this paper for the situation of pupil occlusion. Several experimental results show that the algorithm is robust to locate the position of pupil occlusion and that it can achieve accurate pupil localization.