An audio-based hierarchical smoking behavior detection system based on a smart neckband platform

Smoking behavior detection has attracted much research interest for its significant impact on smokers' physical and mental health. Existing research has shown the potential of using wearable devices for fine-grained smoking puff and session detection by detecting a smoker's content of breathing, lighter usage, breathing, and gesture patterns. However, the existing systems are complex, and they are usually vulnerable to confounding activities and diversity of smoking behavior. To address these limitations, this paper proposes the design and implementation of a simple and compact smart neckband device for smoking detection. The device is equipped with both passive and active acoustic sensors to detect smoking sessions and puffs. We propose a hierarchical processing framework in which the lower-layer detects the sub-movements, i.e., lighter usage, hand-to-mouth gesture and deep breathing, from perceived audio data; and the higher-layer, based on the lower-layer , a´rs detection results, detects smoking puffs and sessions using temporal sequence analysis techniques. Real-world experiments suggest our system can accurately detect smoking puffs and sessions with F1 score of respectively 93.59% and 92.96% in complex environments with the presence of confounding activities and diverse ways of smoking.