Researchers at the University of Cambridge in the U.K., working with colleagues at the Indian Institute of Science, Bangalore and India’s National Institute of Technology, Warangal, have used deep learning to develop a drone surveillance system that automatically detects small groups of people fighting each other.
The system uses computer vision software that runs in real time to detect violent individuals, says the University of Cambridge’s Amarjot Singh.
The researchers trained deep learning algorithms to recognize violent actions by identifying body and limb poses in staged video footage of interns mimicking violence.
Singh replaced some of the neural network layers at the front end with fixed parameters, and used supervised learning toward the back end, exchanging some of the deep learning process with human engineering input. This allowed the resulting ScatterNet Hybrid Deep Learning (SHDL) network to learn more quickly with less data and less available computing power.
The researchers are securing permission from Indian officials to test the system at two upcoming music festivals, and Singh is working to incorporate crowd modeling into the deep learning models.
From IEEE Spectrum
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA