Affect detection in the real world: Recording and processing physiological signals

Abstract

Recording and processing physiological signals from real life for the purpose of affect detection presents many challenges beyond those encountered in the laboratory. Issues such as finding the proper baseline and normalization take on a time dependent meaning. Physical motion also becomes an important factor as these physiological signals often overwhelm those caused by affect. Motion also has an effect on the sensors themselves and precautions must be taken to minimize noise due to changes in placement and loss of connectivity. Ground truth collection is also discussed so that sudden events such as unexpected sounds, bumping into someone in the hallway or having a sneeze are not confused with traumatic affect. In particular, this paper focuses on these issues with respect to recording and processing: galvanic skin response; blood volume pulse; electrocardiogram; electromyogram; respiration and accelerometer signals.