Neural networks application to detect the facts of smoking in video surveillance systems
P V Danilchenko, Nicolae Romanov · Journal of Physics Conference Series · 2021
Abstract The article is devoted to the application of convolutional neural networks and cascade classifiers for tracking smoking facts in relation to video surveillance systems. Two methods of smoking detection are proposed. The first method involves detecting a cigarette against the background of a human face. The second method involves detecting a hand with a cigarette in it. On the basis of the proposed methods, software for a video surveillance system is being developed, which makes it possible to determine the facts of smoking in real time, and thereby prematurely prevent possible undesirable consequences.