Lugano, 25-26 October 2011
Report by Simon Choppin, Agent and sports engineering researcher at Sheffield Hallam University.
Discovering the techniques of Tom Vetter from the University of Basel. His methods (discussed in the report) are reliant on large-scale data collections. Something we have struggled with in the past is how to write streamlined, low-error software for data collections. It struck me that the Software Sustainability Institute could work with researchers who have hardware but need an effective software interface. An efficient and accurate data collection process allows researchers to get more valuable data and gives them more time to use this data and do their research!
Yan Cui from Kaiserslautern University in Germany presented a software-based method of combining scans with the Microsoft Kinect. With this he was able to obtain high-resolution, full-body scans with a single unit. His work is presented in full within the proceedings but there is an onerous implication of programming this from scratch. If Yan Cui was willing to release his software it could form an application of use to a large community engaged with the Kinect.
Accessing a new community with a plethora of impressive algorithms and techniques. While the results of these are astonishing, their use is still restricted due to a lack of knowledge on how they should be used. There is an opportunity here to build guidelines and documentation to inform curious researchers on how to use the many, freely available software solutions already available (which may be impenetrable to those not versed in computer vision processing techniques).
Software and hardware have always been intrinsically linked and this was reflected at the 2nd annual 3D body scanning conference in Lugano. The conference was very well attended by private companies (114 companies opposed to 70 universities in attendance) reflecting the plethora of commercial scanning devices available on the market. Applications of this scanning technology were very varied, with presentations focusing on clinical diagnosis, biomechanical and anthropometric measurement, film making and special effects and (especially) garment design and fitting.
The conference started strongly with a presentation by Professor Tom Vetter from the University of Basel. His research team have used existing statistical methods to create a parametric model of the face which gives control over the bone structure and pigmentation of a modelled face. Their techniques can reconstruct a 3D model face from a single 2D image and, more impressively, can alter that face physically to add or remove weight (for example) while still maintaining the essential structures and physicality of the individual. By showing different face types to user groups and surveying for characteristics such as trustworthiness or sexuality they were able to classify different face types for these properties and apply transformations in order to increase or decrease the trustworthiness of a face, for example. This presentation was a nice example of many of the themes and challenges which were apparent throughout the duration of the conference. For example, when fitting the 3D face model to the 2D image it is essential that the tip of the nose, corners of the eyes and edges of the mouth are all aligned, something Professor Vetter noted as a particular challenge.
This anatomical registration was a recurring discussion point throughout the conference. Gareth Price from the Christie NHS Foundation Trust presented methodologies for locating a patient’s organs from external scans of the abdomen. The company (TC) 2 described the difficulties in taking anatomical measures from disparate, full-body, point cloud data of a consumer. Colin Urquhart discussed the modelling assumptions made when fitting scanned bone data to a 3D scan of a dental patient’s face. After his presentation, I talked to Gareth Price about his work and the wider aims of the Software Sustainability Institute. He currently works with algorithms and software specifically written for their work. I mentioned that the Institute may be able to assist in creating more permanent software solutions which would also be able to benefit other researchers working in similar areas.
What is interesting is that despite many common challenges, software solutions are often specific to the application and would not perform as desired outside of the narrow field to which the technology is applied. Helmut Hungl from the company XYZ RGB showcased an array of incredibly detailed scans of props and inanimate objects for use in the American film industry. However, when these techniques were applied to human scanning, the resolution and quality of scan drops due to movement and other issues not considered when developing the scanning algorithms.
In the past scanning hardware has been relatively bespoke, to which custom software could be afforded. New technology such as the Microsoft Kinect is bringing surface scanning software to a wider audience who are able to apply this technology to new fields which have so far been neglected by the traditional scanning community. These users may not have the experience and knowledge of computer vision and image-processing techniques which is necessary to develop effective, bespoke scanning algorithms.
There exists a clear need for software to provide versatile solutions to the shared scanning challenges faced by a seemingly fragmented community. While open-source libraries are being developed such as Willow Garage’s point cloud library, documentation is disparate and operation currently relies on knowledge of C++. It seems apparent that the community would benefit from guidelines which address the difficulties of dealing with large point cloud data-sets. This area of research will only develop into the future, it is important that as wide a user group as possible benefit from new technology and developments in software.
From a personal perspective we were interested in this conference because we are trying to utilise affordable scanning hardware for biomechanics and human-motion applications. As such it was very valuable to see a different research community and their software approaches. It has also been important to recognise that the impressive results presented by others may use software and techniques which (for whatever reason) are not suitable for our specific approach. We hope to take some of the lessons we are learning to the wider biomechanics community which will inform others which wish to use new scanning technologies in their own work.