Non-intrusive Eating Habits Monitoring Using Millimeter Wave

The potential applications of mmWave technologies are vast and diverse. In this chapter, we aim to explore the feasibility of harnessing millimeter wave technology as an unobtrusive method for eating habits monitoring. The significance of maintaining good health is intrinsically linked to dietary habits. An unhealthy diet can be a precursor to an array of health complications such as obesity, diabetes, heart diseases, and an increased risk of cancers, etc. To aid individuals in keeping track of their dietary behaviors, it is crucial to implement an effective monitoring system. However, conventional sensor-based and camera-based dietary monitoring systems have limitations. They either require users to wear specialized devices, or might raise privacy concerns. Similarly, while WiFi-based methods have demonstrated reasonable performance in specific cases, they have their own set of constraints. Wireless signals often carry environment-specific information, thereby impacting the accuracy of monitoring eating activities. To overcome these challenges, we propose to implement a millimeter wave-enabled eating behavior monitoring system that operates independently of any environment. Our system introduces an innovative approach that mitigates the influence of the surroundings by analyzing mmWave signals in the Doppler-Range domain. This allows us to achieve meticulous monitoring of eating behaviors, facilitated by the construction of a Spatial-Temporal Heatmap through the integration of multiple measurements. Consequently, we can differentiate between dietary activities performed with various utensils such as forks, knives, spoons, chopsticks, or even bare hands. Moreover, we leverage unsupervised learning-based 2D segmentation and an eating period derivation algorithm to accurately estimate the duration of each eating activity. Additionally, our system showcases the potential to infer food categories and determine eating speed. Extensive experiments involving over 1000 eating activities demonstrate the effectiveness of our system, achieving high accuracy in dietary activity recognition with a low false detection rate.