Facial Expression and Gesture Analysis for Emotionally-Rich Man-Machine Interaction - 3D Modeling and Animation: Synthesis and Analysis Techniques for the Human Body - page 193

Table 3. Gesture classification results.

HC-
LF

HC-
HF

LH-
LS

LH-
HS

HoH-
G

HoH-
P

Gesture Class IG

Hand Clapping- Low
Frequency (HC-LF)

5 0 0 0 0 0 0

Hand Clapping- High
Frequency (HC-HF)

0 4 0 0 0 0 1

Lift of the Hand-Low
Speed (LH-LS)

0 0 5 0 0 0 0

Lift of the Hand- High
Speed (LH-HS)

0 0 0 5 0 0 0

Hands over the Head -
Gesture (HoH-G)

0 0 0 0 5 0 0

Hands over the Head -
Posture (HoH-P)

0 0 0 0 0 5 0

Italianate Gestures (IG) 0 1 0 0 0 0 4

Classification Rate
(%)

100 80 100 100 100 100 80

rate of 94.3% was achieved. The experimental results are shown in the
confusion matrix (Table 3).
From the results summarized in Table 3, we observe a mutual misclassification
between “Italianate Gestures” (IG) and “Hand Clapping - High Frequency”
(HC - HF). This is mainly due to the variations on “Italianate Gestures” across
different individuals. Thus, training the HMM classifier on a personalized basis
is anticipated to improve the discrimination between these two classes.

Multimodal Affective Analysis

Facial Expression Analysis Subsystem

The facial expression analysis subsystem is the main part of the presented
system. Gestures are utilized to support the outcome of this subsystem.