Stress Detection in Videos using Machine Learning

Stress is a common occurrence in daily life, many people deal with it at various points of life. Early detection of mental stress can help to prevent many physical problems associated with it. In this study, a technique that would regularly alert candidates based on their emotions and tell them whether they were stressed out or not is suggested. Aweb application determines the stress levels in real-time photos of the user’s face that were captured from live video using a webcam employing a Convolutional Neural Network (CNN) model that can categories facial expressions. The FER-2013 dataset was utilized. The user’s facial location is detected and cropped using a Haar-like feature selection technique and the Viola Jones algorithm. After the face has been scaled and normalized, the facial expressions are next recognized. Then, to determine whether someone is under stress, the emotion with the highest likelihood is utilized. In this method, similar emotions are categorized. To extract facial features and determine the level and value of stress, we employ Dlib’s 68 frontal facial landmarks detector. The accuracy of various models, including Residual Network (ResNet) and Facial Expression Recognition (FER) models, is compared. The highest recognition accuracy was attained by ResNet($98 \%$), followed by CNN (97%), and FER (84%).