Angry or Climbing Stairs? Towards Physiological Emotion Recognition in the Wild

Physiological responses to emotions play a vital role in the field of emotion recognition. Machine-learning models implemented in wristbands or wearables, already exploit unique patterns in physiological responses to provide information about humans emotional states. However, such responses are commonly interfered and overlapped by physical activities, posing a challenge for emotion recognition “in-the-wild”. In this paper, we address this challenge by investigating new features based on the linear regression line and machine-learning models for emotion recognition. We triggered emotions through audio samples and recorded physiological responses from 18 participants before and while performing physical activities. We trained models with the least strenuous physical activity (sitting) and tested with the remaining, more strenuous ones. For three different emotion categories, we achieved classification accuracies up to 67%. Considering individual activities and participants, we achieve up to 73% classification accuracy, indicating the viability of emotion recognition models and features non-sensitive to interferences caused by physical activities.