Computer Will Read Your Mind !
We may soon be interacting with computers in much more unrivalled ways if latest research is anything to go by. This huge possibility was recently brought to fore by researchers at the Robotics Institute of Carnegie Mellon University who developed a mechanism for computers’ interaction and interpretation of human movements from video evidence.
The technology was developed at the Panoptic Studio which houses over 500 video cameras. This revelation means using a laptop computer and camera, human poses can be detected and analyzed, a scenario that enhances human-machine interaction which can be used to foster better communication with computers in more natural ways, for example, using finger gestures or by simply pointing at objects.
Understanding the intricacies of non-verbal cues means robots can now detect actions and moods of people, predict them and make reasonable judgements. With the novel technology, it’s not impossible for self-driving cars to receive early warnings of crossing pedestrians by analyzing the body language and positioning of individuals. The technology also fuels the possibility of improved behavioral diagnosis and treatment of conditions like dyslexia, autism and depression.
There may also be improvements in sports analysis as inch-perfect pose detection by computers will not only help in tracking player positions during play but also permit better analysis of movements involving their legs, arms, heads and general body in real time. The technology can therefore be used for live event analysis or make informed decisions on existing videos.
In driving more research into this technology and its possible applications, researchers involved in the revolutionary development have built a system with a code that can give hand-pose and multi-person estimation. More researchers are giving the technology a shot and with more than 20 commercial groups showing keen interest in licensing it, according to one of the pioneering researchers, there’s surely more to expect of this robust technology in the coming years.
It’s always a difficulty tracking many people at the same time, especially in close contact situations. Thus a technology that only tracks individual pose may not work well when such individuals are in a group, and even more difficult as the group size gets bigger. This problem becomes more obvious when tracking hand motions for example, in which case it is difficult for a camera to see all parts as gestures are made.
This is not unrelated to the fact that unlike the face and other parts of the body, extensive data-sets of laboriously annotated hand images with well-labelled positions and parts do not exist for hand images. To make for this caveat, Yaser Sheikh, associate professor of robotics and his team of researchers created a workaround which involved isolating the legs, faces, arms and other body parts in a scene, before identifying these parts with fixed individuals.
Hanbyul Joo, a robotics researcher stated that many more images of the hand can be taken from other angles after the first image from one part. This made the Panoptic Studio of Carnegie Mellon University an excellent place for the research since up to 500 hand views were retrieved for every single shot. He also informed 31 high-definition cameras powered the generation of a massive data since most cameras were quite off the mark in taking annotated pictures of the hand. The Studio’s capacity to support 3-D models from the current 2-D models of humans will be essential as research intensifies on this dynamic technology.