Human-Centric Interfaces for Ambient Intelligence

According to Hofstede [18], nonverbal behavior is strongly affected by cultural

affordances. The identity dimension, for example, is tightly related to the expression

of emo tions and acceptable emotional displays in a culture, in that, for instance, indi-

vidualistic cultures tolerate individual expressions of anger more easily than do col-

lectivistic cultures. Hofstede et al. [19] explicitly examined the differences that arise

in the use of sound and space for the five dimensions. By relating the results from

their corpus study to Hofstede’s dimensional model, Rehm and colleagues showed

how a user’s expressive gestural behavior can be recognized with high accuracy

and then used to infer the user’s position on Hofstede’s cultural dimensions. With

this contextual information it becomes possible to modify the behavior of an interac-

tive system.

13.6 RECOGNIZING GESTURAL BEHAVIOR FOR HUMAN-CENTRIC

AMBIENT INTELLIGENCE

In Part I of this book, vision-based techniques for gesture recognition were pre-

sented in depth. Here the focus is on input techniques that make use of sensoric

equipment to allow more private interactions. Although vision-based techniques

present the most unobstrusive method for movement analysis and have proven very

successful for recognizing gestural activity (perhap s apart from some minor occlu-

sion problems), they may present a severe threat to privacy in ambient intelligence

environments if the user is unaware of the vision device and does not know which

information is being processed, such as affective state, personality traits, or cultural

background. Thus, obstrusive input methods might be more appropriate for sensi-

tive personal information, as they put the control over information transmitted to

the environment into the hands of the user.

In the remainder of this chapter we present input techniques that make use of

acceleration or physiological sensors like EMG. Such techniques rely on sensors that

are small enough to be carried by the user as handheld devices or to be attached to

her body. It is not unreasonable to assume that such sensors will become integrated

into everyday objects like rings or items of clothing, removing this annoyance of

attaching them altogether.

13.6.1 Acceleration-Based Gesture Recognition

With the advent of Nintendo’s new game console, acceleration-based interactions

have become very popular. Although most commercial games seem to rely on rela-

tively primitive information, like raw acceleration, more sophisticated gesture recog-

nition is possible. Schlo¨mer and colleagues [38] made use of HMMs to analyze

acceleration data. They evaluated their approach with an arbitrary set of five ges-

tures and presented user-dependent recognition rates of up to 93% for this five-class

problem. Rehm and colleagues [36] used acceleration-based recognition to capture

340 CHAPTER 13 Nonsymbolic Gestural Interaction for Ambient Intelligence