This repository contains the code for demonstrating different use cases which leverage facial landmark keypoints. The facial landmark model is a Tf-Js implementation of the paper.
Using the keypoint map we can calculate the distance between keypoints of facial part such as eyelids (upper and lower) and lips (upper and lower) and put a threshold on this distance to identify when the person is drowsing off. A moving average smoothing can reduce noise.