RF Sensing for Smoking Detection at Oil Fields

In this paper, an ultra-wideband (UWB) Radar sensor is used to detect human gestures while smoking or vaping in potentially dangerous areas such as an oil field or a gas station. Existing smoking detection systems are primarily camera-based, which has a number of drawbacks, including poor illumination, training issues with longer video sequence data, and major privacy concerns. The data collected from a UWB Radar is represented in the form of spectrograms. Three classes are considered, namely cigarette, vape and when the subject is not smoking. InceptionV3, VGG19, and VGG16 deep learning algorithms are used to extract spatiotemporal information from the Spectrogram. Finally, by classifying the Spectrograms into the considered gestures, the smoking and/or vaping is accurately identified. The simulation results show that InceptionV3 can achieve a maximum classification accuracy of 90.00%.