Deep Learning-Based Automated Extraction of Anthropometric Measurements From a Single 3-D Scan

• Published in: IEEE transactions on instrumentation and measurement Vol. 70; pp. 1 - 14
• Main Authors: Kaashki, Nastaran Nourbakhsh, Hu, Pengpeng, Munteanu, Adrian
• Format: Journal Article
• Language: English
• Published: New York IEEE 2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Anthropometric measurement; Anthropometry; Coders; Data collection; Deep learning; Deformation; encoder–decoder architectures; Feature extraction; Machine learning; Missing data; Modules; Neural networks; Noise measurement; Noise sensitivity; point cloud; Scanners; template fitting; Three dimensional models; Three-dimensional displays
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2021.3106126

The appearance of 3-D scanners, generating point clouds, has revolutionized anthropometric data collection systems and their applications. Anthropometric data are of paramount importance in several applications, including fashion design, medical diagnosis, and virtual character modeling, all of which require a fully automatic anthropometric measurement extraction method. 3-D-based methods for anthropometric measurement extraction becomes more and more popular due to their improved accuracy compared to classical image-based approaches. Existing 3-D methods can be mainly classified into two categories: landmark and template-based methods. The former is highly dependent on the estimated landmarks which are highly sensitive to noise in the input or missing data. The latter has to iteratively solve an objective function to deform a body template to fit the scan, which is time-consuming while being also sensitive to noise and missing data. In this study, we propose the first approach for automatic contact-less anthropometric measurements extraction based on deep-learning (AM-DL). A novel module dubbed multiscale EdgeConv is proposed to learn local features from point clouds at multiple scales. Multiscale EdgeConv can be directly integrated with other neural networks for various tasks, e.g., classification of point clouds. We exploit this module to design an encoder-decoder architecture that learns to deform a template model to fit a given scan. The measurement values are then calculated on the deformed template model. To evaluate the proposed method, 27 female and 25 male subjects were scanned using a photogrametry-based scanner and measured by an experienced tailor. Experimental results on the synthetic ModelNet40 dataset and on the real scans demonstrate that the proposed method outperforms state-of-the-art methods, and performs sufficiently close to a professional tailor.