Webcam Based Eye Gaze Prediction System with Automatic Calibration for Web Browser

Abstract

Now a day, researches about Eye Gaze Prediction Systems with General Webcam was interested by many researchers and has been continuously developed because it is cheap and can be applied in many ways. Currently, a Browser based open source library for eye gaze prediction has been developed, which is easy to use and can be developed in a variety fields. However, this method is still limited because of low precision and inconveniences because users must perform calibration every time before use it. Therefore, we propose ways to develop and solve the above problems. This research has two objectives. The first is to improve the accuracy of an existing Eye Gaze Prediction System. The solution is using the Simple Moving Average to reduce the volatile of results and increase accuracy. From a results, this method gives a test scores higher than the existing method with statistically significant at 0.01, calculated as 14.96 percent increase from average score of the old method. The second objective is to propose a solution for making the system can recalibrate itself to improve accuracy in a long time without having to perform calibration by the users. We record the gaze data from system in the first 1 minute and then take the recorded data to calculate the boundary of the screen. And then, compare the calculated boundary with the real screen boundary to find error factors. This calculated error factor has been applied back to the next result to increase accuracy. From the test results, we found that over time the average test scores gradually increase. And at the last minute, the average score from this method is higher than the average score from traditional method up to 13.66 percent.