Implementation of MTCNN Facial Feature Extraction on Sleepiness Scale Classification Using CNN

Sleepiness is a condition when the level of human consciousness decreases. Sleepiness is not easy to measure externally. If this is allowed just like that, it would be very dangerous if we were doing activities that requires full control of consciousness such as activities driving. This set of tools and methods for detecting drowsiness has been developed. However, in its implementation the intrusive method uses this tool less practical. In addition, the sleep detection method uses video images as well experiencing problems due to the potential for data loss in the form of facial features. To get this data, you must use a near infrared camera which has a low resolution and lacks detail. Therefore it is necessary an algorithm that is able to detect facial features to the maximum for classifying a person's sleepiness scale. To achieve the goal, a program will be created that functions to perform the extraction facial features and eye features. The program will function to detect number of frames containing facial features and performing condition classification eyes closed or open which will then be saved in a ‘csv’ file to be processed. Furthermore, the data will be carried out in the training process using 1D CNN architecture. The results of the training process that has been carried out The previous model is the model that will be used in performing the scale classification drowsiness. There are 6 experimental scenarios to get the best results. From all the results that have been obtained, it can be concluded that the best model results are the result of the training process with epoch values by 30 and the addition of synthetic data. Accuracy value obtained of 89% and the loss value of 34%.