Ville Rantanen

A survey on the feasibility of surface electromyography (EMG) measurements in facial pacing is presented. Pacing for unilateral facial paralysis consists of the measurement of activity from the healthy side of the face and functional electrical stimulation to reanimate the paralyzed one. The goal of this study is to evaluate the feasibility of surface EMG as a measurement method to detect muscle activations and to determine their intensities. Prior work is discussed, and results from experiments where 12 participants carried out a set of facial movements are presented. EMG was registered from zygomaticus major (smile), orbicularis oris (lip pucker), orbicularis oculi (eye blink), corrugator supercilii (frown), and masseter (chew). Most important facial functions that are limited due to the paralysis are blinking, smiling, and puckering. With majority of the participants, crosstalk between the measured EMG channels was found to be acceptably small to be able to pace smiling and puckering based on detecting their contraction intensities from the healthy side. However, pacing blinking based on orbicularis oculi EMG measurement does not seem possible due to crosstalk from other muscles, but the electro-oculographic (EOG) signals that couple to the same measurement channel could help to detect eye blinks and trigger stimuli. Futhermore, masseter greatly disturbs EMG measurement of most facial muscles, which needs to be addressed in the pacing system to avoid falsely interpreting its activity as the activity of another muscle.

The aim of this paper was to evaluate the use of three facial actions (i.e. frowning, raising the eyebrows, and smiling) in selecting objects on a computer screen when gaze was used for pointing. Dwell time is the most commonly used selection technique in gaze-based interaction, and thus, a dwell time of 400 ms was used as a reference selection technique. A wireless, head-mounted prototype device that carried out eye tracking and contactless, capacitive measurement of facial actions was used for the interaction task. Participants (N=16) performed point-and-select tasks with three pointing distances (i.e. 60, 120 and 240 mm) and three target sizes (i.e. 25, 30 and 40 mm). Task completion times, pointing errors and throughput values based on Fitts' law were used to compare the selection techniques. The participants also rated the techniques with subjective ratings scales. The results showed that the different techniques performed equally well in many respects. However, throughput values varied from 8.38 bits/s (raising the eyebrows) to 15.33 bits/s (smiling) and were comparable to or, in the case of smiling, better than in earlier research with similar interaction techniques. The dwell time was found to be the least accurate selection technique in terms of the magnitudes of point-and-select errors. Smiling technique was rated as more accurate to use than the frowning or the raising techniques. The results give further support for methods that combine facial behavior to eye tracking when interacting with technology.

A wide range of applications can benefit from the measurement of facial activity. The current study presents a method that can be used to detect and classify the movements of different parts of the face and the expressions the movements form. The method is based on capacitive measurement of facial movements. It uses principal component analysis on the measured data to identify active areas of the face in offline analysis, and hierarchical clustering as a basis for classifying the movements offline and in real-time. Experiments involving a set of voluntary facial movements were carried out with 10 participants. The results show that the principal component analysis of the measured data could be applied with almost perfect performance to offline mapping of the vertical location of the facial activity of movements such as raising and lowering eyebrows, opening mouth, raising mouth corners, and lowering mouth corners. The presented classification method also performed very well in classifying the same movements both with the offline and the real-time implementations.