|
ROBOTIC sight and prosthetics which could help the blind to 'see' through their sense of touch are two areas of vision research that may receive a boost from the development of a new, hi-tech virtual reality lab at Oxford University in England. The lab will use a head-mounted display to simulate a virtual world, allowing researchers to study how vision works in ‘the real world’. More importantly, they can 'trick' test subjects, for example, by having the visual scene expand almost imperceptibly as the subject walks through it, permitting the team to study how the vision system reacts. Andrew Glennerster MB BChir, DPhil, Royal Society University Research Fellow and Junior Research Fellow at Queen's College, Oxford, is one of the principal investigators on the team. By performing psychophysical experiments, he hopes to shed light on how the vision represents the perception of a stable, 3-D world. "Normally, as we move around, our eyes jump from object to object about three times a second, yet we are quite unaware of any change. We are also unaware of the swirling patterns of motion which are generated on the retina as we move in a static environment," Dr Glennerster said. He notes that it would be disastrous if we did perceive the dramatic retinal changes produced by saccadic eye movements or the subtler retinal flow produced by head movements. But the question remains: how does the brain make sense of all this rapidly changing visual information? Traditionally, experiments clamped the head in a fixed position to test how the vision system resolved certain stimuli. "Traditional vision research has been undertaken with black and white displays and simple moving stimuli to trying and look at basic (visual perception) mechanisms. With advanced displays (VR) we can create things which are quite like the stimuli in the real world, but we can precisely manipulate them," explains Prof John Wann, Head of Psychology, University of Reading, England. The result should be a better understanding of how vision works in the real world and, the researchers hope, lead to discoveries about the fundamental principles of visual perception. "For example, we're beginning to understand how we make a head-centred representation. For ophthalmologists that is, I'm sure, how they have always been taught about binocular vision. You put the images from each eye together and generate information about 3-D in a co-ordinate frame. The problem is that when you move anywhere, that co-ordinate frame is shot to pieces," Dr Glennerster said. Another beneficiary of this type of research will be the field of robotics. Dr Glennerster is conducting research to develop more robust and detailed hypotheses about how the brain could, in principle, represent and store incoming visual information. Those hypotheses then could be applied to robotic vision. He is working in collaboration with Andrew Fitzgibbon PhD, Royal Society Research Fellow at the Robotics Research Group, Oxford. The group is one of the leading teams in the field of robotic vision. "Their robots must, like humans, store and represent visual information as they move around. One thing is clear, however: the algorithms used by the robots are quite unlike anything the human visual system uses. For example, computer vision systems jump straight from images to a world-based, 3-D co-ordinate representation of the world," Dr Glennerster said. In contrast, there is ample evidence that animals generate some sort of ego-centred representation of space, for example head, body or hand-centred, in addition to, or as an intermediate step towards, the computation of a world-based representation. An understanding of this kind of representation would aid the development of robotic vision akin to human vision. But blindness is the topic that spurs Dr Glennerster's ambition. "What really drives me, what excites me most, is helping blind people to see, but it's a long way off. I would see something like pins on the finger linked to cameras on the (subject's) shoulders. When you think about it, getting signals back from the retina is not that different to getting signals back from the finger. "The signals are only spikes down axons. If people can connect the motor output with the sensory consequence of that output, then they can learn to use it. The key is that we're beginning to think that most vision is based on that kind of a loop," Dr Glennerster said. Because it would be tied to the rules that govern how vision works, the system would theoretically provide the sensory input through, essentially, moving Braille and subjects could learn to navigate ‘visually’ through the 3-D world. It's a hope rooted firmly in the future, however. For now, immersive virtual reality is offering the prospect of new insights into the workings of the vision system. "We can now produce things that start to approximate the natural world quite well. Not perfectly, but the essential components are there. We still haven't got close to a Turing test for VR. No matter how good the display might be, you will still be able to tell the difference compared to the real world. We're nowhere near the level of ‘The Matrix’ and we're not likely to be for at least the next 10 or 20 years. But I think VR is already making a significant contribution to vision science," Prof Wann suggested. |
