University of California, San Diego (UCSD) researchers say they have developed a pedestrian-detection system that performs in near-real time and with higher accuracy compared to conventional systems.
The technology could be used in smart vehicles, robotics, and image and video search systems.
The algorithm combines a traditional computer-vision classification architecture, known as cascade detection, with deep-learning models.
The algorithm first quickly identifies and discards windows that it can easily recognize as not containing a person. The next stages process the windows that are harder for the algorithm to classify. Finally, the algorithm distinguishes between pedestrians and very similar objects.
Although the method is fast, it is not powerful enough when it reaches the final stages because the weak learners used in all the previous stages are identical. The researchers addressed this problem by developing a new algorithm that incorporates deep-learning models in the final stages of a cascaded detector. The new cascade architecture combines classifiers from different families--simple classifiers in the early stages and complex classifiers in the later stages.
"The results we're obtaining with this new algorithm are substantially better for real-time, accurate pedestrian detection," says UCSD professor Nuno Vasconcelos.
Although the algorithm currently only works for binary detection tasks, the researchers aim to extend the cascade technology to simultaneously detect many objects.
From University of California, San Diego
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found