<![CDATA[The movement of the human eye can be useful in various fields, for example in security systems, health, transportation and design interface. In the design interface systems, eye movement used as an interactive system. The system can interact and responses to users by using eye movements. The video-based eye tracking method has the advantage of being practical and convenient during the detection process. This study uses the Convolution Neural Network (CNN) algorithm because it will utilize the advantages of the CNN method to classify and have the most significant results in object recognition. The results of this study indicate that the CNN model that good to use in the classification of eye direction based on facial landmarks is with 2 layers contain 32 filters and 64 filters, batch size 16 in image augmentation with 20 fully connected layers resulting loss value of 0.08, with an accuracy of 0.98 and 8.62 seconds in training time. Test results on videos taken 50 frames randomly three times, resulting in an average accuracy 0.95.]]>
Copyrights © 2020
