Automatic human body segmentation based on feature extraction

Abstract

Purpose – Automatic segmentation of unorganized 3D human body scan data was developed without heuristic specified values. It was reliable in finding the upper body's primary landmarks. The paper aims to discuss these issues. Design/methodology/approach – Quasi boundary point sequence (QBPS) was defined to find the boundary of the human body. Body scan data were categorized by clustering the features extracted from the predefined QBPS. A non-uniform rational B-spline (NURBS) approximation was used to detect the landmarks of the segmented upper torso. Findings – The segmentation method based on feature extraction was reliable regardless of the scan data's fidelity. It was verified that the landmark detection method introduced in this work is more robust than a previous method that utilizes the position of point data. Originality/value – There are several studies of human body segmentation and body landmark detection. This work, however, aims to automate fully segmentation and develop more reliable searching methods. Unlike previous work that uses only 2D human body information, this work uses 3D body information. Furthermore, previous landmark searching methods were superseded by more robust methods applying NURBS approximations.