OpenPose based Smoking Gesture Recognition System using Artificial Neural Network

Abstract

Smoking is an extremely important health problem in modern society. This study focuses on a method for preventing smoking in non-smoking areas, such as public places, as well as the development of an artificial neural network based smoking motion recognition system for more accurately recognizing smokers in such areas. In particular, we attempted to increase the rate of recognition of smoking behaviors using an OpenPose based algorithm and the accuracy of such recognition by additionally applying a hardware device for recognizing cigarette smoke. In addition, a preprocessing method for inputting a dataset into the proposed system is proposed. To improve the recognition performance, four types of dataset models were created, and the most suitable dataset model was selected experimentally. Based on this dataset model, test data were created and input into the proposed neural network based smoking behavior recognition system. In addition, the nearest neighbor interpolation method was selected experimentally as an image interpolation approach and applied to the image preprocessing. When applying experimental data based on learned data, the developed system showed a recognition rate of 70-75%, and the smoking recognition accuracy was increased through the addition of the hardware device.