Database Reference
In-Depth Information
were arranged into 40 classes. Each such motion class contains various different
realizations of the same type of motion, covering a broad spectrum of semantically
meaningful variations.
The skeleton-based mocap file format ASF/AMC was used, which was trans-
formed from the C3D data, the 3D trajectory of the optical markers. The ASF/AMC
data comprises an explicit skeleton structure, providing information about bones,
joints, and the assembly of these basic elements into a skeleton. As a pre-processing
step, the joint locations were normalized by using the hip as the original location
and calculating the location between the joints and the hip. The 3D locations of all
joints were then used as input feature vectors.
The gesture recognition system outlined in Fig.
11.3
has been implemented and
compared in terms of performance to the newly proposed system shown in Fig.
11.8
.
Fig. 11.12
Optical motion capture using retro-reflective markers attached to the actor's body
Search WWH ::
Custom Search