Poster: M-Seven: monitoring smoking event by considering time sequence information via iPhone M7 API

Smartphones are equipped with various sensors that provide rich context information. By leveraging these sensors, several interesting and practical applications have emerged. Accelerometer data has been used, for example, to detect transportation [3], exercise activities [2], etc. A typical approach is to classify activity directly based on features extracted from raw sensing data. Cheng et. al. implemented a different approach by using two-stage classification: the system first detects several sub-behaviors, and uses the combination of attributes to infer higher-level behaviors. Built upon this approach, we foucus on exploring the time sequence of activities, which is an underexplored, yet natural and information-rich indicator. In this work, we explore this time sequence concept through detection of smoking events. In the public area, smoking is usually prohibited. Thus, smokers normally go to outdoor areas with fewer passerbys to smoke. Instead of detecting bio-signals through wearable sensors [1], we leverage movement patterns as indicators; smokers normally start from a stationary state (either the phone is on the desk or in their pocket), walk to the smoking spot which is usually outdoors, stand there for several minutes, then go back to their working area and resume stationary state. Although there are various activities with similar patterns that might cause false positives, e.g., buying lunch from an outdoor food truck, we believe there are subtleties in the sensor data to distinguish them apart, e.g. differences between standing casually (smoking), versus moving periodically when waiting in line (food truck). In this work we demonstrate the detection of the smoking movement pattern through data collected from the primary phone of one smoker for two days.