[Paper Review] NeckSense: A Multi-Sensor Necklace for Detecting Eating Activities in Free-Living Conditions

Shibo Zhang, Yuqi Zhao+5 more|arXiv (Cornell University)|Nov 17, 2019

Eating Disorders and BehaviorsPsychology50 references20 citations

TL;DR

NeckSense is a multi-sensor necklace that unobtrusively detects eating activities in free-living conditions using proximity, ambient light, and IMU sensors to identify chewing sequences and feeding gestures, achieving an F1-score of 81.6% for eating episode detection in exploratory studies and 77.1% in full-day free-living settings, with over 15.8 hours of battery life.

ABSTRACT

We present the design, implementation, and evaluation of a multi-sensor low-power necklace 'NeckSense' for automatically and unobtrusively capturing fine-grained information about an individual's eating activity and eating episodes, across an entire waking-day in a naturalistic setting. The NeckSense fuses and classifies the proximity of the necklace from the chin, the ambient light, the Lean Forward Angle, and the energy signals to determine chewing sequences, a building block of the eating activity. It then clusters the identified chewing sequences to determine eating episodes. We tested NeckSense with 11 obese and 9 non-obese participants across two studies, where we collected more than 470 hours of data in naturalistic setting. Our result demonstrates that NeckSense enables reliable eating-detection for an entire waking-day, even in free-living environments. Overall, our system achieves an F1-score of 81.6% in detecting eating episodes in an exploratory study. Moreover, our system can achieve a F1-score of 77.1% for episodes even in an all-day-around free-living setting. With more than 15.8 hours of battery-life NeckSense will allow researchers and dietitians to better understand natural chewing and eating behaviors, and also enable real-time interventions.

Motivation & Objective

• To develop a low-power, unobtrusive wearable device capable of detecting eating activities throughout an entire waking day in naturalistic environments.
• To address the limitations of existing systems that are either obtrusive, lack long-term validation, or are tested only on homogeneous populations.
• To enable accurate detection of eating episodes across diverse populations, including individuals with and without obesity.
• To support real-time behavioral interventions by enabling continuous, reliable monitoring of chewing and feeding behaviors.
• To create a publicly available, high-quality dataset with video-validated ground truth for future research in eating behavior monitoring.

Proposed method

• The NeckSense necklace integrates a proximity sensor to detect chin distance, an ambient light sensor to detect mouth proximity, and an IMU to capture head motion and lean-forward angles.
• Chewing sequences are identified by fusing features from the proximity, light, and IMU sensors, with classification using a Gradient Boosting Model (GBM).
• Feeding gestures are detected via the lean-forward angle (LFA), which correlates with the act of bringing food to the mouth.
• The system clusters detected chewing sequences into distinct eating episodes based on temporal proximity and behavioral consistency.
• Sensor data is processed in real time to minimize power consumption, enabling over 15.8 hours of continuous operation on a single charge.
• Validation is performed using a shoulder-mounted camera and self-reported meal logs, with ground truth established by professional labeling of video data.

Experimental results

Research questions

• RQ1 Can a multi-sensor necklace reliably detect chewing sequences and eating episodes in free-living conditions across diverse individuals?
• RQ2 How does the performance of NeckSense vary in long-term, all-day monitoring compared to shorter, intermittent studies?
• RQ3 To what extent does the system generalize across individuals with varying body mass indices, including obese and non-obese participants?
• RQ4 What is the impact of sensor fusion (proximity, light, IMU) on detection accuracy compared to single-sensor approaches?
• RQ5 Can the system maintain high accuracy while operating for over 15 hours on a single battery charge in real-world settings?

Key findings

• NeckSense achieved an F1-score of 81.6% for eating episode detection in an exploratory, semi-free-living study with selective meal capture.
• In a fully free-living, all-day study, the system maintained a robust F1-score of 77.1% for eating episode detection.
• The system detected chewing sequences with an F1-score of 73.7% at the per-second level, demonstrating high accuracy even at fine-grained temporal resolution.
• NeckSense successfully monitored participants for an average of 15.8 hours on a single charge, enabling full-day, continuous data collection.
• The dataset collected from 20 participants over 2 weeks includes diverse eating scenarios—such as eating in cars, while slouching, or talking—providing rich, real-world behavioral data.
• The inclusion of both obese and non-obese participants demonstrated the system’s generalizability across diverse body types and eating behaviors.