Statistical prediction of body landmark locations on surface scans
Authors: Matthew P. Reed, Byoung-Keon Park, K. Han Kim, Monica L.H. Jones.
Automatic location of body landmarks on human three - dimensional surface scan data can improve the efficiency of anthropometric studies. Previous methods have relied on local features and heuristics to identify landmarks. This paper presents a purely statistical technique based on a statistical body shape model (SBSM). An optimization technique is used to identify a vector of principal component scores that maps the SBSM to the scan data. The locations of 98 landmarks and joint centers were predicted by this method and compared to the manually extracted locations for 213 men scanned in a relaxed standing posture. The mean (across landmarks) median error was 19 mm. Considering only the torso and head, the mean median error was 14 mm. Joint location discrepancies were generally less than 25 mm, comparing favourably with outer methods for locating joint centers.