Researchers at Carnegie Mellon University (CMU) have developed an improved method for coding important context from an image that enables a significant advance in detecting tiny faces.
When applied to benchmarked datasets of faces, the new method reduced errors by a factor of two, and 81% of the faces found using the CMU methods proved to be actual faces, compared with up to 64% for prior methods.
The researchers say the technique could have multiple applications, such as performing headcounts to calculate the size of a crowd.
In addition, detecting small items in general will become increasingly important as self-driving cars must monitor and evaluate traffic conditions in the distance.
The new method uses "foveal descriptors" to encode context in a way similar to how human vision is structured.
The researchers also improved the ability to detect tiny objects by training separate detectors for different scales of objects.
From Carnegie Mellon News (PA)
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA
No entries found