Skip to main content Start main content

Image-based Precise 3D Human Modelling

Just take two photos for accurate body measurements within seconds


Singles’ Day on 11 November started out in 1990s as an anti-Valentine’s festival celebrating singlehood.  But over the years, it has become an eagerly anticipated sales frenzy for all online shoppers in China.  On Singles’ Day 2018, e-commerce sales hit record high at RMB 213.5 billion[1].  However, research also shows that up to 30% of products sold online are returned mostly because of wrong sizes[2], incurring immense administration cost, not to mention the shipping cost on returns that many vendors have to bear and the detriment on customer satisfaction.  To facilitate better size recommendation for online shoppers and to make online orders of bespoke tailor-made clothing possible, Dr Tracy Mok, Associate Professor, and Dr Shuaiyin Zhu, PhD graduate, Institute of Textiles and Clothing, developed an Image-based Precise 3D Human Modelling technology that projects measurements of over 50 different body parts of a person based on two photos only.  The estimated measurements can achieve accuracy within 1 centimetre.

Mechanism

Traditionally, non-contact shape modelling uses either scan-based, image-based or example-based techniques.  Scanning body forms involve large-scale high-precision scanners that are expensive and bulky.  Asking shoppers to have their body scanned before buying anything online is just impractical.  Conventional image-based or example-based technique uses too much approximation so that the estimated measurements are not accurate enough to be useful.

The 3D modelling system developed by Dr Mok and Dr Zhu, on the other hand, uses a different logic.  “The system doesn’t scan the whole body as one 3D object, but breaks it down into more than 50 different local features.  The local 2D shapes are analysed to re-construct local 3D forms, which are then compared with over 10,000 actual human scans via big data analytics before being put together as one.  It is similar to an MRI or a CT scan – a person is divided into slices and each slice is measured separately,” explained Dr Mok.

A user has to punch in his height and weight first.  Then he has full-length photos of his front and his side taken with the app.  And his measurements will be generated within 10 seconds.  “The app and online application programme interface (API) are designed to separate the human body from complex background, and to compensate for any lens distortion or perspective.  It also reads the data of the gyroscope and accelerometer in a smartphone so that it can direct the user to hold the phone at a certain angle to avoid errors,” added Dr Mok.  When the user takes the photos wearing form-fitting clothes, the projected measurements deviate only within 1 centimetre from their actual figures.  Such high accuracy allows tailored clothing or even intimate apparels that fit perfectly to be made without the need to take measurements in person.

Commercialization

To foster entrepreneurship among students and to encourage the applications of research results to benefit the people at large, PolyU partnered with Shanghai Technology Entrepreneurship Foundation for Graduates (STEFG) to establish STEFG-PolyU China Entrepreneurship Fund, which financially supported TOZI, a start-up based in Shenzhen founded by Dr Zhu and his partners.  TOZI licensed the 3D modelling technology and released the free app "1measure" that features a simplified version of the system.  There are currently five apparel retailers in China that use the system for size recommendation, virtual try-ons or made-to-order clothing.

In December 2018, the Image-based Precise 3D Human Modelling system won a Silver Medal at the 1st Asia Exhibition of Inventions Hong Kong.

 

[1] Kharpal, Arjun. “Alibaba sets new Singles Day record with more than $30.8 billion in sales in 24 hours.” 11 Nov 2018. CNBC. Retrieved at https://www.cnbc.com/2018/11/11/alibaba-singles-day-2018-record-sales-on-largest-shopping-event-day.html

[2] Datayx. “网购退货率超三成 用户数据分析成电商出路” [Online sales return rate exceeds 30pc; user data analytics a way out for vendors.] 9 March 2018. Tencent Cloud Trade Column. Retrieved at https://cloud.tencent.com/developer/article/1056748

Your browser is not the latest version. If you continue to browse our website, Some pages may not function properly.

You are recommended to upgrade to a newer version or switch to a different browser. A list of the web browsers that we support can be found here