This study presents an algorithm that can detect people's facial features being studied and then applied mainly on daily basis activities, as an example in driving which is detection of driver drowsiness. In this study, the algorithm named 'PERCLOS' which stands for 'percentage of eye closure' was tested to detect face by using two face landmark detectors, that are pre-Trained model and library Dlib's 68-points facial landmark and 468 3D face landmarks detector from MediaPipe by Google as an alternative and detects the condition of a person's eye based on Eye Aspect Ratio (EAR). Initial assessment of the Dlib's solution on 151,537 frames (about 84 minutes) of one of tested subjects revealed that 98.66 of eye states were properly identified, resulting in 378 blinks to be recorded. Despite having rather good accuracy, the algorithm produced 166 more blinks than the 212 blinks that were expected. As for MediaPipe, with 264 blinks and only 52 additional blinks, the MediaPipe Face Mesh solution was able to categorize the identical subject with a classification accuracy of 99.87. Additionally, adaptive thresholds for different subjects were applied in order to investigate a way to improve the studied algorithm. Surprisingly, the adaptive threshold method being studied resulted in decreasing accuracy and precision for some of the subjects. For one of tested subject, the resulted precision of studied algorithm somehow drops from 100 to 98.60. © 2022 IEEE.