State-of-Art Facial Monitoring: A Machine Learning Perspective

This study proposes the use of real-time human facial tracking for interacting with robots. One of the key elements that enable humanoid robots to do difficult tasks is their visual system. As a result, there have been many investigations on gaze stability artificially, of which a large number are based on reflexes in the biological visual system. The gaze-stabilization approach incorporates Using processing of images, kinematic inverse, and control via feedback, the required gaze behavior is produced. The system is divided into two components. The human face is recognized, and positioned in respect to the original image, and its proportions are created for use in the second part in the first section. The second component determines the location and dimensions of the face by moving the camera by the offset between the intervals while monitoring the face. In contrast to prior work, which employed the Haar cascade method to recognize human faces, in plenty of situations, the Kanade-Lucas-Tomasi (KLT) technique is applied for face tracking. The offset between image frames, which was obtained from the prior stage, is used to offset the camera. Optimized deep neural networks like ResNet can raise the accuracy and performance of face recognition and real-time tracking techniques like Kalman filters are used to keep face tracking constant between frames. The findings demonstrate that real-time monitoring of human faces was still successful even when individuals were using face-side postures, glasses, or hats. This article describes the most frequent problems with face tracking, position changes, variations in face resolution, variations in lighting, and facial distortion.