Search

Liquid identification plays an essential role in our daily lives. However, existing RF sensing approaches still require dedicated hardware such as RFID readers and UWB transceivers, which are not readily available to most users. In this article, we propose Akte-Liquid, which leverages the speaker on smartphones to transmit acoustic signals, and the microphone on smartphones to receive reflected signals to identify liquid types and analyze the liquid concentration. Our work arises from the acoustic intrinsic impedance property of liquids, in that different liquids have different intrinsic impedance, causing reflected acoustic signals of liquids to differ. Then, we discover that the amplitude-frequency feature of reflected signals may be utilized to represent the liquid feature. With this insight, we propose new mechanisms to eliminate the interference caused by hardware and multi-path propagation effects to extract the liquid features. In addition, we design a new Siamese network-based structure with a specific training sample selection mechanism to reconstruct the extracted feature to container-irrelevant features. Our experimental evaluations demonstrate that Akte-Liquid is able to distinguish 20 types of liquids at a higher accuracy, and to identify food additives and measure protein concentration in the artificial urine with a 92.3% accuracy under 1 mg/100 mL as well.

Earphones are emerging as the most popular wearable devices and there has been a growing trend in bringing intelligence to earphones. Previous efforts include adding extra sensors (e.g., accelerometer and gyroscope) or peripheral hardware to make earphones smart. These methods are usually complex in design and also incur additional cost. In this paper, we present Earmonitor, a low-cost system that uses the in-ear earphones to achieve sensing purposes. The basic idea behind Earmonitor is that each person's ear canal varies in size and shape. We therefore can extract the unique features from the ear canal-reflected signals to depict the personalized differences in ear canal geometry. Furthermore, we discover that the signal variations are also affected by the fine-grained physiological activities. We can therefore further detect the subtle heartbeat from the ear canal reflections. Experiments show that Earmonitor can achieve up to 96.4% Balanced Accuracy (BAC) and low False Acceptance Rate (FAR) for user identification on a large-scale data of 120 subjects. For heartbeat monitoring, without any training, we propose signal processing schemes to achieve high sensing accuracy even in the most challenging scenarios when the target is walking or running.

Gesture detection based on radio frequency signals has gained increasing popularity in recent years due to several benefits it has brought, such as eliminating the need to carry additional devices and providing better privacy. In traditional methods, significant breakthroughs have been made to improve recognition accuracy and scene robustness, but the limited computing power of edge devices (the first-level equipment to receive signals) and the requirement of fast response for detection have not been adequately addressed. In this article, we propose a lightweight Wi-Fi gesture recognition system, referred to as WiFine, which is designed and implemented for deployment on low-end edge devices without the use of any additional high-performance services in the process. Toward these goals, we first design algorithms for phase difference selection and amplitude enhancement, respectively, to tackle the problem of data drift caused by user change. Then, we design a cross-dimension fusion method to extract features of finer granularity from information of different dimensions, thus solving the precision problem of feature granularity. Finally, we design a lightweight neural network architecture by leveraging redundancy to reduce computational cost while ensuring satisfactory recognition accuracy. Extensive experimental results show that the proposed system achieves fast recognition of various actions with an accuracy up to 96.03% in 0.19 seconds.

Contactless RF-based sensing techniques are emerging as a viable means for building gesture recognition systems. While promising, existing RF-based gesture solutions have poor generalization ability when targeting new users, environments or device deployment. They also often require multiple pairs of transceivers and a large number of training samples for each target domain. These limitations either lead to poor cross-domain performance or incur a huge labor cost, hindering their practical adoption. This paper introduces Wi-Learner, a novel RF-based sensing solution that relies on just one pair of transceivers but can deliver accurate cross-domain gesture recognition using just one data sample per gesture for a target user, environment or device setup. Wi-Learner achieves this by first capturing the gesture-induced Doppler frequency shift (DFS) from noisy measurements using carefully designed signal processing schemes. It then employs a convolution neural network-based autoencoder to extract the low-dimensional features to be fed into a downstream model for gesture recognition. Wi-Learner introduces a novel meta-learner to "teach" the neural network to learn effectively from a small set of data points, allowing the base model to quickly adapt to a new domain using just one training sample. By so doing, we reduce the overhead of training data collection and allow a sensing system to adapt to the change of the deployed environment. We evaluate Wi-Learner by applying it to gesture recognition using the Widar 3.0 dataset. Extensive experiments demonstrate Wi-Learner is highly efficient and has a good generalization ability, by delivering an accuracy of 93.2% and 74.2% - 94.9% for in-domain and cross-domain using just one sample per gesture, respectively.

Radio-Frequency (RF) based human activity recognition (HAR) enables many attractive applications such as smart home, health monitoring, and virtual reality (VR). Among multiple RF sensors, mmWave radar is emerging as a new trend due to its fine-grained sensing capability. However, laborious data collection and labeling processes are required when employing a radar-based sensing system in a new environment. To this end, we propose RF-CM, a general cross-modal human activity recognition framework. The key enabler is to leverage the knowledge learned from a massive WiFi dataset to build a radar-based HAR system with limited radar samples. It can significantly reduce the overhead of training data collection. In addition, RF-CM can work well regardless of the deployment setups of WiFi and mmWave radar, such as performing environments, users' characteristics, and device deployment. RF-CM achieves this by first capturing the activity-related variation patterns through data processing schemes. It then employs a convolution neural network-based feature extraction module to extract the high-dimensional features to be fed into the activity recognition module. Finally, RF-CM takes the generalization knowledge from WiFi networks as guide labels to supervise the training of the radar model, thus enabling a few-shot radar-based HAR system. We evaluate RF-CM by applying it to two HAR applications, fine-grained American sign language recognition (WiFi-cross-radar) and coarse-grained gesture recognition (WiFi-cross-RFID). The accuracy improvement of over 10% in both applications demonstrates the effectiveness of RF-CM. This cross-modal ability allows RF-CM to support more cross-modal applications.

This paper focuses on a fundamental question in Wi-Fi-based gesture recognition: "Can we use the knowledge learned from some users to perform gesture recognition for others?". This problem is also known as cross-target recognition. It arises in many practical deployments of Wi-Fi-based gesture recognition where it is prohibitively expensive to collect training data from every single user. We present CrossGR, a low-cost cross-target gesture recognition system. As a departure from existing approaches, CrossGR does not require prior knowledge (such as who is currently performing a gesture) of the target user. Instead, CrossGR employs a deep neural network to extract user-agnostic but gesture-related Wi-Fi signal characteristics to perform gesture recognition. To provide sufficient training data to build an effective deep learning model, CrossGR employs a generative adversarial network to automatically generate many synthetic training data from a small set of real-world examples collected from a small number of users. Such a strategy allows CrossGR to minimize the user involvement and the associated cost in collecting training examples for building an accurate gesture recognition system. We evaluate CrossGR by applying it to perform gesture recognition across 10 users and 15 gestures. Experimental results show that CrossGR achieves an accuracy of over 82.6% (up to 99.75%). We demonstrate that CrossGR delivers comparable recognition accuracy, but uses an order of magnitude less training samples collected from the end-users when compared to state-of-the-art recognition systems.

Drowsiness detection is critical to driver safety, considering thousands of deaths caused by drowsy driving annually. Professional equipment is capable of providing high detection accuracy, but the high cost limits their applications in practice. The use of mobile devices such as smart watches and smart phones holds the promise of providing a more convenient, practical, non-invasive method for drowsiness detection. In this article, we propose a real-time driver drowsiness detection system based on mobile devices, referred to as dWatch, which combines physiological measurements with motion states of a driver to achieve high detection accuracy and low power consumption. Specifically, based on heart rate measurements, we design different methods for calculating heart rate variability (HRV) and sensing yawn actions, respectively, which are combined with steering wheel motion features extracted from motion sensors for drowsiness detection. We also design a driving posture detection algorithm to control the operation of the heart rate sensor to reduce system power consumption. Extensive experimental results show that the proposed system achieves a detection accuracy up to 97.1% and reduces energy consumption by 33%.

Wireless sensing builds upon machine learning shows encouraging results. However, adopting wireless sensing as a large-scale solution remains challenging as experiences from deployments have shown the performance of a machine-learned model to suffer when there are changes in the environment, e.g., when furniture is moved or when other objects are added or removed from the environment. We present Rise, a novel solution for enhancing the robustness and performance of learning-based wireless sensing techniques against such changes during a deployment. Rise combines probability and statistical assessments together with anomaly detection to identify samples that are likely to be misclassified and uses feedback on these samples to update a deployed wireless sensing model. We validate Rise through extensive empirical benchmarks by considering 11 representative sensing methods covering a broad range of wireless sensing tasks. Our results show that Rise can identify 92.3% of misclassifications on average. We showcase how Rise can be combined with incremental learning to help wireless sensing models retain their performance against dynamic changes in the operating environment to reduce the maintenance cost, paving the way for learning-based wireless sensing to become capable of supporting long-term monitoring in complex everyday environments.

International audience; Respiration rate sensing plays a critical role in elderly care and patient monitoring. The latest research has explored the possibility of employing Wi-Fi signals for respiration sensing without attaching a device to the target. A critical issue with these solutions includes that good monitoring performance could only be achieved at certain locations within the sensing range, while the performance could be quite poor at other "dead zones." In addition, due to the contactless nature, it is challenging to monitor multiple targets simultaneously as the reflected signals are often mixed together. In this work, we present our system, named LungTrack, hosted on commodity RFID devices for respiration monitoring. Our system retrieves subtle signal fluctuations at the receiver caused by chest displacement during respiration without need for attaching any devices to the target. It addresses the dead-zone issue and enables simultaneous monitoring of two human targets by employing one RFID reader and carefully positioned multiple RFID tags, using an optimization technique. Comprehensive experiments demonstrate that LungTrack can achieve a respiration monitoring accuracy of greater than 98% for a single target at all sensing locations (within 1 st − 5 th Fresnel zones) using just one RFID reader and five tags, when the target's orientation is known a priori. For the challenging scenario involve two human targets, LungTrack is able to achieve greater than 93% accuracy when the targets are separated by at least 10 cm.

Wireless sensor networks (WSNs) have been used for many long-term monitoring applications with the strip topology that is ubiquitous in the real-world deployment, such as pipeline monitoring, water quality monitoring, vehicle monitoring, and Great Wall monitoring. The efficiency of routing strategy has been playing a key role in serving such monitoring applications. In this article, we first present a robust geographic opportunistic routing (GOR) approach—LIght Propagation Selection (LIPS)—that can provide a short path with low energy consumption, communication overhead, and packet loss. To overcome the complication caused by the multi-turning point structure, we propose the virtual Plane mirror (VPM) algorithm, inspired by the light propagation, which is to map the strip topology into the straight one logically. We then select partial neighbors as the candidates to avoid blindly involving all next-hop neighbors and ensure the data transmission along the correct direction. Two implementation problems of VPM—transmission spread angle and the communication range—are thoroughly analyzed based on the percolation theory. Based on the preceding candidate selection algorithms, we propose a GOR algorithm in the strip topology network. By theoretical analysis and extensive simulation, we illustrate the validity and higher transmission performance of LIPS in strip WSNs. In addition, we have proved that the length of the path in LIPS is two times the length of the shortest path via geometrical analysis. Simulation results show that the transmission success rate of our approach is 26.37% higher than the state-of-the-art approach, and the communication overhead and energy consumption rate are 33.11% and 40.23% lower, respectively.

As one of the best ways to learn and appreciate the Chinese culture, Chinese calligraphy is widely practiced and learned all over the world. Traditional calligraphy learners spend a great amount of time imitating the image templates of reputed calligraphers. In this paper, we propose an RF-based Chinese calligraphy template, named RF-Copybook, to precisely monitor the writing process of the learner and provide detail instructions to improve the learner's imitating behavior. With two RFID tags attached on the brush pen and three antennas equipped at the commercial RFID reader side, RF-Copybook tracks the pen's 3-dimensional movements precisely. The key intuition behind RF-Copybook's idea is that: (i) when there is only direct path signal between the tag and the antenna, the phase measured at the reader changes linearly with the distance, (ii) the reader offers very fine-grained phase readings, thus a millimeter level accuracy of antenna-tag distance can be obtained, (iii) by combing multiple antenna-tag distances, we can quantify the writing process with stroke based feature models. Extensive experiments show that RF-Copybook is robust against the environmental noise and achieves high accuracies across different environments in the estimation of the brush pen's elevation angle, nib's moving speed and position.

Many emerging applications and the ubiquitous wireless signals have accelerated the development of Device Free localization (DFL) techniques, which can localize objects without the need to carry any wireless devices. Most traditional DFL methods have a main drawback that as the pre-obtained Received Signal Strength (RSS) measurements (i.e., fingerprint) in one area cannot be directly applied to the new area for localization, and the calibration process of each area will result in the human effort exhausting problem. In this paper, we propose FALE, a fine-grained transferring DFL method that can adaptively work in different areas with little human effort and low energy consumption. FALE employs a rigorously designed transferring function to transfer the fingerprint into a projected space, and reuse it across different areas, thus greatly reduce the human effort. On the other hand, FALE can reduce the data volume and energy consumption by taking advantage of the compressive sensing (CS) theory. Extensive real-word experimental results also illustrate the effectiveness of FALE.