Researchers from North Carolina State University have developed a new technique that improves the capability of computer vision technologies to better identify and different objects in an image, a process called segmentation.
Computer vision technologies utilize algorithms to section or outline the objects, in a picture. For instance, dividing the outline of a pedestrian against the backdrop of a busy street.
These algorithms rely on specified parameters - programmed values - to segment images. By way of example, if there is a shift in color that crosses a specific threshold, a computer vision program will interpret it as a dividing line between two objects.
But there is a challenge here. Even tiny changes in a parameter may result in very different computer vision success. For instance, if a person crossing the street walks in and from shady locations, that would influence the color a pc sees - and the monitor may then "see" the person disappearing and reappearing, or translate the person and the shadow as a single, large object such as a car.
"Some algorithm parameters might work better than others in any particular set of conditions, and we desired to know how to combine multiple parameters and algorithms to create better picture segmentation by computer imagery applications," says Edgar Lobaton, an assistant professor of electrical and computer engineering at NC State and senior author of a paper on the job.
Lobaton and Ph.D. student Qian Ge developed a technique that compiles segmentation information from multiple algorithms and aggregates them, developing a brand new version of the picture. This new image is then flashed again, dependent on how consistent any given section is across all the original input algorithms.
"Visually, the outcomes of the technique look better than any given algorithm on its own," Lobaton says. "But the character of this work does not line up with the present metrics for quantifying computer vision precision. So we need to come up with a new means of assessing computer vision accuracy - that's a potential job for us"
Lobaton notes the new picture segmenting technique can be used in real time, processing 30 frames each second. This is due, in part, to the fact that most of the computational steps could be run in parallel, rather than sequentially.
|