Researchers have come up with an algorithm that provides improved measurement and analysis of clothed human body shapes, with potential applications both in virtual fitting rooms at apparel stores and in the security field.

Posture analysis helps re-create clothed human body shapes

High-street fashion and apparel retailers are increasingly installing ‘virtual fitting rooms’ for their customers. Of course if the system is to provide really accurate clothes fittings it has to be able to assess all the different individual body shapes and posture variants – which presents clothes store management with a conundrum. While it is relatively easy to produce a digital version of any given piece of clothing, it has proved much more difficult to come up with a genuine three-dimensional visualisation of the actual customer fitting. And such details are extremely important in a business that depends heavily on the personal shopping experience. Now a German-Canadian team of researchers have been working to create a system capable of bridging this gap by simulating individual body shape, in relation to the subject’s posture, while wearing clothes.

Body shape and posture

Existing analysis systems provide a means to roughly estimate the size and shape of a person captured on camera by reducing the thickness of the different layers of clothing worn. However, the German-Canadian team set out to integrate movement into the calculations. Using a camera to track motion sequences, such as when the subject is walking towards the camera, they have developed a model that can analyse and learn about these various types of movement. Their system is thus ‘trained’ to combine static analysis of the human form with posture analysis that takes account of the way the clothes drape the body in action. From this multiple analysis can be created an image that corresponds closely to an individual person’s body form, filtering out as far as possible the apparent changes in volume due to the undulations of the clothes worn.   The result of this work is an algorithm capable of calculating human size and form, which can work either by analysing, frame by frame, pictures from a single camera angle or by recording images from multiple viewpoints.

Virtual nudity

Using this algorithm, the researchers have shown that they can simulate very closely an accurate picture of a person’s body form. This innovation could well prove extremely useful at clothing stores as the virtual changing room should now be able to provide the customer not only with a highly-realistic virtual try-on for fit but also a view of how different cuts and fabric volumes would flow and look on the body.  We can picture a customer inside the booth, trying on a dress, shirt or suit, simulating various movements and seeing how they affect the fit and line of the apparel. Meanwhile, there are also some potentially important applications of the algorithm in the security field, such as detecting unusual shapes underneath a person’s clothing.

By Quentin Capelle