An Embedded Computer Vision Method to Extract Percentage of Eye Close for Detecting Drowsiness of a Safety Driving System

Various factors can cause accidents, but the main factor that dominates the causes of accidents is the driver's actions, especially continuing to drive in a state of drowsiness. To avoid accidents, a safety driving system is needed to inform the driver when in a drowsiness condition. This paper reports on the early stages of developing a safety driving system implemented in an embedded computer vision method. We calculated the perclos from the ear and the ear from eye landmarks. We obtained significant results from the perclos when the driver had driven for 3 hours. The average perclos for 3 hours is 0.152, while after more than 3 hours driving is 0.590. This result is significant in distinguishing the driver's condition, especially in developing rules for a safety driving system. The processing speed we obtained in extracting eye landmarks was 189.91 milliseconds at a speed of 10 fps. This speed is fast enough to detect drowsiness. Furthermore, developing a drowsiness detection system will involve a professional driver subject who works as a transporter and adding psychological signal characteristics such as ECG signal and driving behavior modality parameters in producing a multimodal based decision-making system.