Eyeball Movement Detection Using Sector Division Approach and Extreme Learning Machine

Eyeball movement is being widely used for many purposes. A lot of research is trying to find the best approaches and methods to detect, track and recognize the movements. In this research, we propose an approach to detect the direction of eyeball movements using Sector Division Approach and Extreme Learning Machine (ELM). The extraction process of Sector Division is detecting facial image, detecting the eye location using subset points in the Facial Landmark. Selected eye location is segmented and through several processes such as image cropping, conversion into grayscale image, blurring process, and finally binary process. The final image in the binary process is divided into 9 (nine) sectors and extracted resulting in 9 feature vectors. ELM is used to classify the eyeball movement. The optimal number of hidden neurons identified first before the model is used in the testing step. A total of 50 data is used to train the ELM to classify the eyeball movement. The ELM model is executed 5 (five) times to reduce the variability of the random weight in the ELM model. Testing is done by evaluating each eyeball movement using 12 still images in each direction. Based on the experiment, a number of 20 hidden neurons results in the highest predictive accuracy and is used in the testing step. The result shows that the proposed model is able to achieve a satisfactory result by showing an accuracy of 81.67%. The result of this study could be beneficial to be used in similar studies as using a small number of training data, basic feature extraction, and a small number of feature vectors could achieve satisfactory accuracy.