Facial Landmarks Detection System with OpenCV Mediapipe and Python using Optical Flow (Active) Approach

To achieve positive detection results, a variety of face landmark methods supported by the convolutional neural network have been developed. The instability landmarks thus emerge in video frames as a result of CNNs, on the other hand, are extremely sensitive to input picture noise. This paper provides a light and effective face landmark identification technology based on a lightweight U-Net model based on semantic segmentation and an Optical Flow (Active) Approach (OFA) for solving the problem of landmark shaking. The OFA employs a quick optical flow approach to determine the motion path of the landmark, as well as a route to increasing landmark maintenance. A lightweight U-Net model is used to predict face landmarks with a reduced size of the model and lower computational. To subsume unstable shaking, the predictable face landmarks are given into the OFA technique as well. Finally, various benchmark datasets are used to produce a comparison of many common methodologies as well as the proposed detection process. A lightweight U-Net model is used to model face landmarks in reduced model size and lower computational. To subsume the unstable shaking, the predicted face landmarks are given into the OFA technique as well. Finally, various benchmark datasets are used to produce a comparison of many common methodologies as well as the proposed detection process.