In an effort to automate early diagnosis of autism, researchers are testing a new system that combines Kinect motion sensors with computer-vision algorithms capable of detecting telltale signs of the behavioral disorder.
The experiment is underway at the University of Minnesota’s Institute of Child Development in Minneapolis. Inside the institute, five Kinect depth-sensing camera rigs installed in the nursery are used to monitor groups of roughly 10 children between the ages of 3 and 5 years old as they play. The New Scientist reports:
The cameras identify and track children based on their shape and the colour of the clothes they are wearing. The information is fed to three PCs, which run software that logs each child’s activity level – including how they move each of their limbs – and plots it against the room’s average. The system can flag up children who are hyperactive or unusually still – both possible markers for autism.
Medical staff can then decide whether the child requires closer attention from a specialist for a one-on-one diagnosis
Ultimately, the team hopes to merge the Kinect work with another project it is working on. By studying video footage of children interacting with a psychiatrist, computer-vision algorithms learn to identify behavioural markers as designated on the Autism Observation Scale for Infants (.pdf). The system measures traits like a child’s ability to follow an object as it passes in front of the eyes, as well as noting certain mannerisms or postures that are classified as being early signs of a possible [autism spectrum disorder].
Researchers say that automating diagnosis of autism could result in children beginning speech therapy sooner and getting help learning social and communication skills, which can have a significant impact on treatment outcomes. Long-term, researchers hope to develop a video game for Kinect capable of testing a child as they played with a parent, meanwhile identifying any concerns.