Gaze tracking plays a crucial role in human-computer interaction, psychological assessment, virtual reality, and other related fields. The key step in gaze tracking is to accurately locate the pupil under different environmental conditions. The accuracy of pupil localization can be compromised by partial reflection, eyelid or eyelash occlusion, and other similar factors. In this paper, we present an algorithm that utilizes high and low thresholds and pupil compensation to address the aforementioned issues. Firstly, we preprocess the original eye image by applying Gaussian blur and downsampling techniques, locate the darkest pixel area, and determine the high and low thresholds. Next, we extract the pupil contour using binarization and Canny edge detection. Finally, we refine the pupil contour to subpixel level and fit an ellipse to the contour using the least-squares method to obtain the pupil parameter information. Additionally, we propose a method for removing occlusions and compensating contour points, which allows for accurate pupil localization under varying degrees of occlusion. The experimental results show that the algorithm in this paper has a positioning accuracy of 70.63% within the 5-pixel error range on the LPW dataset, which is better than some algorithms.