Multimodal biosignal sensor data handling for emotion recognition

We present an experimental setup, sensor data handling, and evaluation framework for emotion recognition, based on multimodal biosignal sensor data. For labeled data acquisition we developed an emotion elicitation block, with a bank of labeled videos containing different triggering stimuli. A biosignal acquisition apparatus was used to collect multimodal data, namely: Electromyography (EMG); Electrocardiography (ECG); Electrodermal Activity (EDA); Blood Volume Pulse (BVP); Peripheral Temperature (SKT); and Respiration (RESP). An automated biosignal processing and feature extraction toolbox was developed to convert raw data into meaningful parameters. Experimental results revealed trends associated with triggering events, providing a baseline for emotion recognition. Through LOOCV with a k-NN classifier, we obtained recognition rates of 81% to distinguish between positive and negative emotions, and of 70% to distinguish between positive, neutral, and negative emotions.