A Deep CNN Framework for Distress Detection Using Facial Expression

Recent medical developments have projected panic attacks as a forerunner in high-stress job environments. This medical cause has emerged to a considerable extent due to the masses’ lifestyle and food consumption habits. Assessment and prevention of such attacks have become imperative to arrest the situation. In this regard, the present study attempts to detect human distress from facial expressions. Convolutional Neural Networks (CNN) serves as one of the best feature extractors. AlexNet being one of the primitive CNN models, has been employed to study the stress content in a facial expression.AlexNet can perform multi-GPU training, which widely reduces the training time for larger models. Training other models comparatively require higher computations that result in escalated time and energy, which might cause consequent lesser efficiency. We have achieved a training accuracy of 93.4%, and validation accuracy of 92.5% —the image set comprised 35340 images generated from 593 video sequences from 123 people at 30fps. Although AlexNet being one of the primitive CNN models, the results of this study are motivating.