Winter Olympics: the new video technology that could help power Britain’s skeleton team to gold

Skeleton is an exhilarating Winter Olympic sport in which athletes race head-first down an ice track at speeds reaching over 80 miles per hour (130km/h). While the event can look basic at first glance, success relies heavily on highly engineered equipment and extensive wind‑tunnel testing – much like elite Olympic track cycling programmes.

Each run begins with the athlete pushing a sled (also known as a “tea tray”) explosively off the starting block, then sprinting rapidly for about 30 metres downhill. After diving on the sled, they ride the rest of the course with their head just a few inches above the ice. The sleds have no brakes, and riders wear only a thin suit and helmet for protection.

A powerful start is considered the defining component of skeleton performance. So, developing a skeleton athlete’s strength and power while refining their pushing technique is a central focus in the lead-up to competitions. The biggest of all these, the Winter Olympics, is being held in Milan and Cortina d'Ampezzo, Italy, this month. Skeleton events start on February 12.

While Britain does not tend to rank highly in Winter Olympic sports, in skeleton it has won a world-best nine Olympic medals, including three golds. Over the past ten years, my colleagues and I at the University of Bath have worked with Team GB skeleton athletes to help improve their starts, using a form of “markerless” motion capture technology.

But the applications of this technology extend far beyond the Winter Olympics. There is potential for it to replace traditional motion capture systems in the film, TV and gaming industries, and to be used in injury rehabilitation.

How motion analysis began

The origins of motion analysis can be traced back to the pioneering work of English photographer Eadweard Muybridge in the late 19th century. Muybridge developed early techniques for capturing sequences of images, including documenting equine gait.

By manually annotating specific features across successive images, researchers have since been able to build a detailed picture of how a person or animal moves. But while this method was the standard for many decades, it was both time- and labour-intensive.

So, technological advances in cameras and computer processing led to the development of automated methods of motion analysis – notably, marker-based motion capture. This uses reflective markers placed on key parts of the body, which are automatically tracked by infra-red cameras as the person moves around.

In film, animation and gaming, this mean an actor’s body movements and facial expressions can be translated into to realistic CGI characters. Marker-based technology is currently the most widely used 3D motion analysis technique across the film, gaming and health sectors, with an estimated global market value of over US$300 million (£220 million).

However, this advanced technology has limitations too, including the need for specialist equipment, controlled laboratory environments, and lengthy preparation time to attach the markers. These can be problematic in sports and many other fields – particularly during live competitions and public performances.

As a result, the field of motion analysis has come almost full circle. Thanks to major advances in computer vision and artificial intelligence, biomechanists such like me are once again extracting detailed movement information directly from video images – but this time in an automated way.

The markerless motion capture systems we use rely on deep‑learning models that are trained on a huge number of images of people performing everyday activities. When applied to unseen images, the algorithms can then automatically detect the same body landmarks. By fusing multiple camera views, a simplified digital 3D skeleton can be extracted, from which the person’s movement across time can be modelled and analysed.

Analysing the optimum technique

Markerless motion capture makes it possible to unobtrusively measure athletes’ movements outside the lab, in training and even during competitions. Our recent research has demonstrated its value in many different sports, including badminton, tennis and Olympic weightlifting.

In skeleton, the unique, bent-over position at the start of each run, as the athlete sprints alongside the sled with one hand holding it, makes this form of biomechanical analysis particularly important.

Using markerless motion capture, we have explored the differing roles of an athlete’s limbs in the push-start performance, comparing these biomechanics with conventional sprinting. Importantly, we have also validated this markerless approach by comparing it with a traditional marker‑based system.

The optimum starting technique for each skeleton athlete is shaped by their physical characteristics, including factors such as relative limb lengths and flexibility. Analysing each athlete’s pushing technique, how it relates to their performance and how this evolves over time, can help give them a crucial competitive edge during this all-important first phase of each skeleton run.

Medals can be won and lost by hundredths of seconds as athletes sprint away from the starting block. In these first few seconds, we hope Britain’s athletes reap the benefit of our markerless motion capture technology.

Steffi Colyer receives funding via the Centre for the Analysis of Motion, Entertainment Research & Applications (CAMERA) from the UKRI's Engineering and Physical Sciences Research Council.