Estimation of Human Pinnae Measurements through Computer Vision
Mentor:Kenneth J Faller II, Assistant Professor, Computer Engineering Program, California State University Fullerton
Head-Related Transfer Functions (HRTFs) are digital filters used to create three-dimensional (3D) binaural (two-channel) spatial audio. HRTFs are anatomically dependent and vary greatly from one individual to another. The two most common types of HRTFs are individualized HRTFs and generic HRTFs. Individualized HRTFs are individually measured using specialized equipment, which is expensive and cumbersome, removing their availability from the general public. Generic HRTFs are measured on KEMAR mannequins that have average anatomical features which, in theory, would be sufficient for individuals with close-to-average anatomical features. Although, the generic HRTFs reduce cost and increase the availability of HRTFs, they usually result in increased localization errors (e.g., increased front-to-back reversals). To increase HRTFs availability to the general public and to reduce localization errors, many researchers have developed methods of generating customizable HRTFs. One method exists that generates customizable HRTFs based on the anthropometric measurements of a potential user’s outer-ear or pinnae [K. Faller, 2009]. Currently, the only anthropometric measurement estimation method known to the authors is to obtain the measurements manually. This current research focuses on the estimation of anthropometric measurements from a digital image of a potential user’s pinnae by applying computer vision techniques. An initial step is to detect the individual structures of the pinnae. The first, and most crucial, structure successfully detected is the concha which has been shown to be crucial for human sound source localization [J. Blauert, Spatial Hearing, 1996]. The next step in this research is to automatically estimate the dimensions (height and width) of the concha. The results of this work will be applied to estimate other anthropometric measurements of the pinnae (e.g., dimensions of the ear, fossa, helix, etc.). Additionally, to reduce computational complexity of the resulting estimation technique, statistical estimation of a portion of the anthropometric measurements will be explored.