Robot vision lags behind human sight
By pitting human vision against that of machines for the first time, computer scientists have shown that machines still struggle with interpreting visual patterns, compared to their human counterparts.
The study, published in Proceedings of the National Academy of Sciences today, implies that computer vision research still has a long way to go before machines have visual perception rivalling that of humans.
Visual machine learning has applications in a variety of fields, ranging from space exploration to homeland security, industrial robotics and vehicle safety. Anything that requires a machine to ‘see’ or extract useful information from an image in real-time encompasses computer vision technology.
Machines are far superior to humans at certain tasks - such as airport security surveillance and facial recognition - but they struggle with more abstract concepts like categorising objects.
For example, if you show a machine a picture of a house, it cannot identify a chimney, roof or window by their shape and relative spatial arrangement. This study shows exactly how far behind machines are in being able to detect individual shapes and describe categories.
“Humans understand and characterise images better, but statistics and computers are more powerful,” said study author and robotics engineer François Fleuret from the Idiap Research Institute in Martigny, Switzerland. “By combining these, we can reach a whole new level of performance.”