We address the open problem of reliably detecting oral health behaviors passively from wrist-worn inertial sensors. mORAL detects brushing and flossing behaviors, without the use of instrumented toothbrushes so that the model is applicable to brushing with still prevalent manual toothbrushes. We show that for detecting rare daily events such as toothbrushing, adopting a model that is based on identifying candidate windows based on events, rather than fixed-length timeblocks, leads to significantly higher performance.
- Trained and tested on 2,797 hours of sensor data
- Collected over 192 days on 25 participants (using video annotations for ground truth labels)
- Brushing model achieves 100% median recall with a false positive rate of one event in every nine days of sensor wearing
- Average error in estimating the start/end times of the detected event is 4.1% of the interval of the actual toothbrushing event
The original dataset that used for the mORAL paper contains data from 25 participants (12 males, 13 females; mean age 28.5 ± 7.6 years), each identified with a random identifier, who completed one-week-long brushing and flossing study. Data from three out of 25 participants are excluded from this released dataset since they did not consent to release their data.
Devices and Sensors
- MotionSense (“wristband”) that included 3-axis accelerometers sampled at 16 Hz and 3-axis gyroscopes sampled at 32 Hz
- Data collected via a wireless Bluetooth connection to a study smartphone platform
- Samsung Galaxy S5 smartphone platform
- Bluetooth-enabled Oral-B toothbrush (“SmartBrush”)
Details & Specifications
mORAL Dataset Description: A Dataset for Inferring Oral Hygiene Behaviors in-the-wild using wrist-worn intertial sensors.
Hours of Sensor Data
Data was collected from 25 users of everyday devices: sensors from smartphone and smartwatches.
Days of Data Collection
Data was collected from users that were engaged in their regular natural behavior continuously for more than 7 days for each user.