 |  |
| Qian ZHANG |  |
|
 |
|
Tencent Professor of Engineering, Chair Professor of Department of CSE, HKUST FIEEE, Changjiang Chair Professor, Huazhong University of Science and Technology (2012-2015)
Co-Director and founder, Huawei-HKUST Innovation Lab Director, Digital Life Research Center, HKUST
My Citations (Google Scholar click here)
Department of Computer Science and Engineering Hong Kong University of Science and Technology Clear Water Bay, Kowloon Hong Kong
Email: qianzh@cse.ust.hk URL: http://www.cs.ust.hk/~qianzh Office: Room 3533 (via Lift 25-26), Academic Building Tel: 852-23588766 Fax: 852-23581477
|
Communication as sensing service --- from wearable sensing to contactless sensing
The past decade has witnessed the surge of interest in daily activity monitoring. Human-centered sensing attempts to have a comprehensive assessment of people’s living habits, health condition and even mental state through monitoring their sleeping quality, dietary information, exercise intensity, daily routines and etc. Numerous applications could benefit from the advanced human sensing systems including elder caring in smart hospitals and homes, dietary management for the diabetics, fitness guidance for the white-collars and etc, which enhances the social medical services, facilitates the health self-management and helps improve the physique of the entire people.
However, there is still a long way to bring the effective, portable, unobtrusive, affordable and comprehensive human sensing system to the general public, which has attracted considerable attentions and efforts from both academia and industry. We envisioned the great potential and social benefits of human-centered sensing many years ago. One major research of our group in HKUST is to build reliable and practical human sensing systems by leveraging wearables and ubiquitous ambient signals, including the following aspects:
Sleep Monitoring
Sleep quality is an important health metric to human beings. Sleep disorders can lead to severe health problems such as heart attack, high blood pressure and stroke. However, current gold standard for sleep monitoring is Polysomnograph (PSG) which requires the patient to sleep in a specific sleep center with a bunch of sensors attached to his/her body. Not only is this method costly and cumbersome, but it may also lead to unrepresentative result as the patient may feel uncomfortable and nervous with a lot of sensors. As a result, many patients remain undiagnosed and miss the suitable time for treatment. To fill the gap, our group has built two portable and convenient instruments for sleeping monitoring at home.
-
Real-time Automatic SleepScoring system(RASS) is their first research outcome, which is a portable sleep monitoring system which can detect sleep stages and occurrence of sleep apnea in real time. It only needs to attach one probe to the user’s finger, which greatly reduces its impact on user’s sleep. It extracts representative features from the blood oxygen, photoplethysmogram (PPG) and actigraph collected by the probe and leverages fuzzy directed graph support vector machine (FDGSVM) to classify the sleep stages. The novel algorithm design enables the system to estimate sleep stages in real time. They tested RASS on 48 subjects and achieved an accuracy exceeding 84%.
Dietary
Monitoring
Dietary monitoring can provide valuable information for disease diagnosis, body weight control, and dietary habit management, and thus it is welcomed by patients, dieters, and nutritionists. Many existing solutions either require tedious manual recording or may impede normal daily activities. To bring practical dietary monitoring into daily use, Prof. Zhang’s group has designed two dietary monitoring systems, an on-body approach (a pair of diet-aware glasses) and an off-body approach (a set of smart utensils).
-
The diet-aware glass is a wearable, unobtrusive and reliable dietary monitoring system, which does not require users’ manual logging and has no invasion of privacy. The key idea is that when people wear glasses, the temples of the glasses are in touch with the lower part of the temporalis muscle, one of the mastication muscles. By integrating an electromyography (EMG) sensor into glasses, the glasses can measure the muscle activity of the temporalis to detect intake-related events. Specifically, the research group used adaptive thresholding for chewing spotting from low-quality EMG signals. Besides, they extracted effective features to distinguish mastication from other similar activities, such as talking and laughing. Furthermore, they proposed a real-time algorithm to only keep and transmit the data containing potential durations of food intake, which greatly saves the wearable device’s battery life and storage space. Experiment results show that the proposed system can achieve 96% accuracy for detecting chewing cycles and up to 90.8% accuracy for classifying five types of known food.
-
To further boost the capacity of dietary management systems on tracking meal composition, the research group proposed SmartU, a new utensil design that can recognize meal composition during the intake process, without user intervention or on-body instruments. Smart-U makes use of the fact that light spectra reflected by foods are dependent on the food ingredients. By analyzing the reflected light spectra, Smart-U can recognize what food is on top of the utensil. They proposed effective schemes to address the ambient light interference and minimize disturbance of LEDs to users’ eyes. More importantly, they designed special lighting patterns for LEDs and built machine learning models to predict food category and nutrients. They built two prototypes of Smart-U, a spoon and a glass, and conducted extensive experiments to test the performance of Smart-U. It can recognize up to 20 types of food with 93% accuracy and can work robustly under different temperatures, lighting conditions, and when in motion. They also took the primarily attempt to predict nutrition information in milk and recognize mixed foods. It is believed that Smart-U moves a significant step toward automatic dietary monitoring that enables people to track their meal composition.
Prof. Zhang and her group members have devoted much efforts in wearable-based human sensing and have successfully built several systems. Now, they are working on a more ambitious envision where more ubiquitous, unobtrusive and contactless sensing is enabled to penetrate into people’s daily lives without them even being aware of the sensing modalities. Toward this goal, they are currently conducting researches on human-centered sensing by leveraging ubiquitous ambient signals. They envision that the wireless transceivers, such as Wi-Fi routers and millimeter-wave radars can be utilized to monitor the people’s locations, daily activities, vital signs or even infer their emotion state. The feasibility lies in the fact that human activities will affect and modulate the wireless signals in the environment, and different activities will have different impacts on the signals. By analyzing the subtle differences of the received wireless signals, it is possible to recognize different kinds of activities. Advanced learning techniques are expected to be integrated into the sensing system for performance boosting. Such contactless sensing scheme frees the users of wearing any device, which could benefit plenty of applications, such as smart homes that can automatically adjust lighting and air conditioning according to people’s emotion, and smart devices that can assess people’s sleep quality without attaching any sensor to their bodies. Prof. Zhang and her group members are actively exploring this research area and trying to build reliable and practical systems to realize this vision.