Extraction of Anthropometric Measures from 3D-Meshes for the Individualization of Head-Related Transfer Functions

Dinakaran M, Grosche P, Brinkmann F, Weinzierl S. Extraction of Anthropometric Measures from 3D-Meshes for the Individualization of Head-Related Transfer Functions. In: AES Convention 140. Audio Engineering Society; 2016. Paper 9579. Available from: https://aes.org/publications/elibrary-page/?id=18277

@inproceedings{Dinakaran2016_18277,
  author = {Dinakaran, Manoj and Grosche, Peter and Brinkmann, Fabian and Weinzierl, Stefan},
  title = {{Extraction of Anthropometric Measures from 3D-Meshes for the Individualization of Head-Related Transfer Functions}},
  booktitle = {AES Convention 140},
  note = {Paper 9579},
  year = {2016},
  month = may,
  publisher = {Audio Engineering Society},
  url = {https://aes.org/publications/elibrary-page/?id=18277}
}

TY  - CPAPER
TI  - Extraction of Anthropometric Measures from 3D-Meshes for the Individualization of Head-Related Transfer Functions
AU  - Dinakaran, Manoj
AU  - Grosche, Peter
AU  - Brinkmann, Fabian
AU  - Weinzierl, Stefan
T2  - AES Convention 140
M1  - Paper 9579
PY  - 2016
DA  - 2016/05/06
UR  - https://aes.org/publications/elibrary-page/?id=18277
PB  - Audio Engineering Society
LA  - en
AB  - Anthropometric measures are used for individualizing head-related transfer functions (HRTFs) for example, by selecting best match HRTFs from a large library or by manipulating HRTF with respect to anthropometrics. Within this process, an accurate extraction of anthropometric measures is crucial as small changes may already influence the individualization. Anthropometrics can be measured in many different ways, e.g., from pictures or anthropometers. However, these approaches tend to be inaccurate. Therefore, we propose to use Kinect for generating individual 3D head-and-shoulder meshes from which anthropometrics are automatically extracted. This is achieved by identifying and measuring distances between characteristics points on the outline of each mesh and was found to yield accurate and reliable estimates of corresponding features. In our experiment, a large set of anthropometric measures was automatically extracted for 61 subjects and evaluated in terms of a cross-validation against the manually extracted correspondent.
ER  -