Deepfakes have emerged as a significant threat to online communications, enabling nearly indistinguishable impersonation of executives, public figures, and trusted contacts during video calls. While state-of-the-art deepfake detection models can achieve high accuracy offline, deploying them in real-time video conferencing systems remains challenging: the added computation quickly violates interactive latency budgets and greatly limits scalability. Our empirical analysis reveals that video decoding and frame movement dominate the detection pipeline, together accounting for approximately 86.6% of per-frame processing time.

Every year, the U.S. economy loses more than${\$}$411 billion because of work performance reduction, injuries, and traffic accidents caused by microsleep. To mitigate microsleep's consequences, an unobtrusive, reliable, and socially acceptable microsleep detection solution throughout the day, every day is required. Unfortunately, existing solutions do not meet these requirements. In this paper, we propose WAKE, a novel behind-the-ear wearable device for microsleep detection. By monitoring biosignals from the brain, eye movements, facial muscle contractions, and sweat gland activities from behind the user's ears, WAKE can detect microsleep with a high temporal resolution. We introduce a Three-fold Cascaded Amplifying (3CA) technique to tame the motion artifacts and environmental noises for capturing high fidelity signals. Through our prototyping, we show that WAKE can suppress motion and environmental noise in real-time by 9.74-19.47 dB while walking, driving, or staying in different environments, ensuring that the biosignals are captured reliably. We evaluated WAKE using gold-standard devices on 19 sleep-deprived and narcoleptic subjects. The Leave-One-Subject-Out Cross-Validation results show the feasibility of WAKE in microsleep detection on an unseen subject with average precision and recall of 76 and 85 percent, respectively.

While the global healthcare market of wearable devices has been growing significantly in recent years and is predicted to reach $60 billion by 2028, many important healthcare applications such as seizure monitoring, drowsiness detection, etc. have not been deployed due to the limited battery lifetime, slow response rate, and inadequate biosignal quality.

Microsleep, caused by sleep deprivation, sleep apnea, and narcolepsy, costs the U.S.'s economy more than $411 billion/year because of work performance reduction, injuries, and traffic accidents. Mitigating microsleep's consequences require an unobtrusive, reliable, and socially acceptable microsleep detection solution throughout the day, every day. Unfortunately, existing solutions do not meet these requirements.

Over 50 million people undergo surgeries each year in the United States, with over 70% of them filling opioid prescriptions within one week of the surgery. Due to the highly addictive nature of these opiates, a post-surgical window is a crucial time for pain management to ensure accurate prescription of opioids. Drug prescription nowadays relies primarily on self-reported pain levels to determine the frequency and dosage of pain drug. Patient pain self-reports are, however, influenced by subjective pain tolerance, memories of past painful episodes, current context, and the patient's integrity in reporting their pain level. Therefore, objective measures of pain are needed to better inform pain management.

We propose a compact noninvasive glucose monitoring system using polarized light, where a user simply needs to place her palm on the device for measuring her current glucose concentration level. The primary innovation of our system is the ability to minimize light scattering from the skin and extract weak changes in light polarization to estimate glucose concentration, all using low-cost hardware. Our system exploits multiple wavelengths and light intensity levels to mitigate the effect of user diversity and confounding factors (e.g., collagen and elastin in the dermis). It then infers glucose concentration using a generic learning model, thus no additional calibration is needed. We design and fabricate a compact (17 cm x 10 cm x 5 cm) and low-cost (i.e., <$250) prototype using off-the-shelf hardware. We evaluate our system with 41 diabetic patients and 9 healthy participants. In comparison to a continuous glucose monitor approved by U.S. Food and Drug Administration (FDA), 89% of our results are within zone A (clinically accurate) of the Clarke Error Grid. The absolute relative difference (ARD) is 10%. The r and p values of the Pearson correlation coefficients between our predicted glucose concentration and reference glucose concentration are 0.91 and 1.6 x 10-143, respectively. These errors are comparable with FDA-approved glucose sensors, which achieve ≈90% clinical accuracy with a 10% mean ARD.

Drones have carried weapons, drugs, explosives and illegal packages in the recent past, raising strong concerns from public authorities. While existing drone monitoring systems only focus on detecting drone presence, localizing or fingerprinting the drone, there is a lack of a solution for estimating the additional load carried by a drone. In this paper, we present a novel passive RF system, namely DroneScale, to monitor the wireless signals transmitted by commercial drones and then confirm their models and loads. Our key technical contribution is a proposed technique to passively capture vibration at high resolution (i.e., 1Hz vibration) from afar, which was not possible before. We prototype DroneScale using COTS RF components and illustrate that it can monitor the body vibration of a drone at the targeted resolution. In addition, we develop learning algorithms to extract the physical vibration of the drone from the transmitted signal to infer the model of a drone and the load carried by it. We evaluate the DroneScale system using 5 different drone models, which carry external loads of up to 400g. The experimental results show that the system is able to estimate the external load of a drone with an average accuracy of 96.27%. We also analyze the sensitivity of the system with different load placements with respect to the drone's body, flight modes, and distances up to 200 meters.

Patients with respiratory diseases require frequent and accurate blood oxygen level monitoring. Existing techniques, however, either need a dedicated hardware or fail to predict low saturation levels. To fill in this gap, we propose a phone-based oxygen level estimation system, called PhO 2 , using camera and flashlight functions that are readily available on today’s off-the-shelf smartphones. Since the phone’s camera and flashlight were not made for this purpose, utilizing them for oxygen level estimation poses many difficulties. We introduce a cost-effective add-on together with a set of algorithms for spatial and spectral optical signal modulation to amplify the optical signal of interest while minimizing noise. A near-field-based pressure detection and feedback mechanism are also proposed to mitigate the negative impacts of user’s behavior during the measurement. We also derive a non-linear referencing model with an outlier removal technique that allows PhO 2 to accurately estimate the oxygen level from color intensity ratios produced by the smartphone’s camera. An evaluation on COTS smartphone with six subjects shows that PhO 2 can estimate the oxygen saturation within 3.5% error rate comparing to FDA-approved gold standard pulse oximetry. In addition, our evaluation in hospitals presents high correlation with ground-truth qualified by the 0.83/1.0 Kendall τ coefficient.

Frequent blood pressure (BP) assessment is key to the diagnosis and treatment of many severe diseases, such as heart failure, kidney failure, hypertension, and hemodialysis. Current "gold-standard'' BP measurement techniques require the complete blockage of blood flow, which causes discomfort and disruption to normal activity when the assessment is done repetitively and frequently. Unfortunately, patients with hypertension or hemodialysis often have to get their BP measured every 15 minutes for a duration of 4-5 hours or more. The discomfort of wearing a cumbersome and limited mobility device affects their normal activities. In this work, we propose a device called eBP to measure BP from inside the user's ear aiming to minimize the measurement's impact on users' normal activities while maximizing its comfort level. eBP has 3 key components: (1) a light-based pulse sensor attached on an inflatable pipe that goes inside the ear, (2) a digital air pump with a fine controller, and (3) a BP estimation algorithm. In contrast to existing devices, eBP introduces a novel technique that eliminates the need to block the blood flow inside the ear, which alleviates the user's discomfort. We prototyped eBP custom hardware and software and evaluated the system through a comparative study on 35 subjects. The study shows that eBP obtains the average error of 1.8 mmHg and -3.1 mmHg and a standard deviation error of 7.2 mmHg and 7.9 mmHg for systolic (high-pressure value) and diastolic (low-pressure value), respectively. These errors are around the acceptable margins regulated by the FDA's AAMI protocol, which allows mean errors of up to 5 mmHg and a standard deviation of up to 8 mmHg.

This paper explores a new wearable system, called TYTH, that enables a novel form of human computer interaction based on the relative location and interaction between the user's tongue and teeth. TYTH allows its user to interact with a computing system by tapping on their teeth. This form of interaction is analogous to using a finger to type on a keypad except that the tongue substitutes for the finger and the teeth for the keyboard. We study the neurological and anatomical structures of the tongue to design TYTH so that the obtrusiveness and social awkwardness caused by the wearable is minimized while maximizing its accuracy and sensing sensitivity. From behind the user's ears, TYTH senses the brain signals and muscle signals that control tongue movement sent from the brain and captures the miniature skin surface deformation caused by tongue movement. We model the relationship between tongue movement and the signals recorded, from which a tongue localization technique and tongue-teeth tapping detection technique are derived. Through a prototyping implementation and an evaluation with 15 subjects, we show that TYTH can be used as a form of hands-free human computer interaction with 88.61% detection rate and promising adoption rate by users.

We present CapBand, a battery-free hand gesture recognition wearable in the form of a wristband. The key challenges in creating such a system are (1) to sense useful hand gestures at ultra-low power so that the device can be powered by the limited energy harvestable from the surrounding environment and (2) to make the system work reliably without requiring training every time a user puts on the wristband. We present successive capacitance sensing, an ultra-low power sensing technique, to capture small skin deformations due to muscle and tendon movements on the user's wrist, which corresponds to specific groups of wrist muscles representing the gestures being performed. We build a wrist muscles-to-gesture model, based on which we develop a hand gesture classification method using both motion and static features. To eliminate the need for per-usage training, we propose a kernel-based on-wrist localization technique to detect the CapBand's position on the user's wrist. We prototype CapBand with a custom-designed capacitance sensor array on two flexible circuits driven by a custom-built electronic board, a heterogeneous material-made, deformable silicone band, and a custom-built energy harvesting and management module. Evaluations on 20 subjects show 95.0% accuracy of gesture recognition when recognizing 15 different hand gestures and 95.3% accuracy of on-wrist localization.

Accurately measuring and monitoring patient's blood oxygen level plays a critical role in today's clinical diagnosis and healthcare practices. Existing techniques however either require a dedicated hardware or produce inaccurate measurements. To fill in this gap, we propose a phone-based oxygen level estimation system, called PhO2, using camera and flashlight functions that are readily available on today's off-the-shelf smart phones. Since phone's camera and flashlight are not made for this purpose, utilizing them for oxygen level estimation poses many challenges. We introduce a cost-effective add-on together with a set of algorithms for spatial and spectral optical signal modulation to amplify the optical signal of interest while minimizing noise. A light-based pressure detection algorithm and feedback mechanism are also proposed to mitigate the negative impacts of user's behavior during the measurement. We also derive a non-linear referencing model that allows PhO2 to estimate the oxygen level from color intensity ratios produced by smartphone's camera.