Pain Level Recognition via Biophysiological Signals and Facial Expressions using Machine Learning

Abstract

In this thesis we present a pain level recognition system which built on pipeline. Pain is a displeasure feeling that informs us about threats to our bodies and the process of measuring pain can be challenging since it has very different and complex characteristics and based on multimodal features which have been extracted from the individual. Proposed here are the different sub-models that are built to solve these complications, we suggest systems for pain level recognition. A state-of-the-art in machine learning techniques have been applied to classify the pain level using recorded videos for facial expressions of the individuals and the biophysiological signals (EMG, ECG, SCL ,...). In this research, we applied a feature extraction for videos by iterating through the frames, after that we extracted five groups of features from each bio-physiological signal (amplitude, frequency, stationarity, entropy, linearity). we proposed that a multimodal majority voting technique consisting of three learned models could improve the efficiency of our system. By using the trained models using the video signals, bio-physiological signals and the fusion of the two signals, the experiments have proved that the performance of the majority voting improved the performance of each individual model on its own.