University of Windsor teaching computers to see like 'human eye'
A researcher at the University of Windsor is working on a more affordable way to use technology for a computer to see and identify objects that would then help build things like self-driving cars, better surveillance and more advanced medical imaging.
Dibyendu Mukherjee, a recent PhD graduate from the school's electrical engineering program, has been capturing video and still images.
He puts this data into a computer, which segments the images. The computer is trained to figure out what type of objects are in the segmented video or image using the algorithm.
"The computer itself cannot actually process any image or video it can simply have a camera added to it and it can give the pictures," explains Mukherjee. "After it gets the pictures out, algorithms can process them and notify the computer of what are the objects that are inside that video or image."
Useful for traffic, medical imaging, surveillance
Mukherjee said the technology would be especially useful if cameras were mounted above traffic signals.
In the future the real-time images from the cameras and sensors could be used to calculate risk of an accident and even notify drivers, he said.
It could also be used in the medical field to help identify different diseases and illnesses.
"High-tech cameras can also take pictures of the eye or other human body parts that can be analyzed for damaged tissues," said Mukherjee. "The computer can be trained by our algorithms to identify and differentiate between a normal tissue structure and a damaged one. This could mean avoiding surgery or catching a disease while it's still early enough to cure."
Knowing that this research could have usable results is what keeps Mukherjee motivated.
"When you say, 'I'm simply working with a computer' that doesn't mean much, it doesn't motivate you, but when when you know that your research is going to be used by the people, that is the main motivation of our research."
How it works, more in depth
The process works in real time. Mukherjee said that's where it gets tricky because the computer has to identify moving objects and people as events take place.
Mukherjee uses high-end stereo cameras that take two images at a time to help build a three-dimensional image. He also uses range cameras that use infrared light to take pictures rendering depth information.
He now wants to perfect a computer's ability to identify objects, like the human eye, by using lower-end cameras and cheaper instruments to cut down the cost.
"The basic goal of all our work is to simplify processes and that starts with finding cheaper instruments," he said. "I captured video using a laptop camera, the kind you'd use for Skyping, and it was successful. Our robust algorithm still works with those cheaper cameras."
He said the more affordable technology could be on the market at some point this year, or next year.
