Yinhuan Dong

Facial expressions are a crucial aspect of non-verbal communication and often reflect underlying emotional states. Researchers often use facial emotion detection as a tool to gain insights into cognitive processes, emotional states and cognitive load. The conventional camera-based methods to sense human emotions are privacy intrusive, lack adaptability, and are sensitive to variability. These technologies have limited generalization and may not adapt well to variations in ambient lighting, facial landmark localization, facial occlusions and emotion intensity. Radio Frequency (RF) sensing offers promising avenues for improvement with contactless, non-invasive, privacy-preserving and reliable radar-based measurements. The proposed framework utilizes deep-learning techniques to classify facial micro-doppler signatures, generated from an ultra-wideband (UWB) radar. The method relies on continuous multi-level feature learning from radar time-frequency Doppler measurements. The spatiotemporal facial features are extracted from the radar data to train deep learning models. The proposed system achieves a high multiclass classification accuracy of 77% on the continuous streamed data covering basic emotions of anger, disgust, fear, happy, neutral and sadness. The system can transform next-generation multi-modal hearing aids with emotion-aware listening effort and cognitive load detection. This can be particularly useful in translating the emotion-assisted cognitive effort for real-time speech enhancement and personalized auditory experience.

In recent years, crowdsourcing approaches have been proposed to record the WiFi signals annotated with the location of the reference points (RPs) extracted from the trajectories of common users to reduce the burden of constructing a fingerprint (FP) database for indoor positioning. However, crowdsourced data is usually sensitive to crowd density. The positioning accuracy degrades in some areas due to a lack of FPs or visitors. To improve the positioning performance, this paper proposes a scalable WiFi FP augmentation method with two major modules: virtual reference point generation (VRPG) and spatial WiFi signal modeling (SWSM). A globally self-adaptive (GS) and a locally self-adaptive (LS) approach are proposed in VRPG to determine the potential unsurveyed RPs. A multivariate Gaussian process regression (MGPR) model is designed to estimate the joint distribution of all WiFi signals and predicts the signals on unsurveyed RPs to generate more FPs. Evaluations are conducted on an open-source crowdsourced WiFi FP dataset based on a multi-floor building. The results show that combining GS and MGPR can improve the positioning accuracy by 5% to 20% from the benchmark, but with halved computation complexity compared to the conventional augmentation approach. Moreover, combining LS and MGPR can sharply reduce 90% of the computation complexity against the conventional approach while still providing moderate improvement in positioning accuracy from the benchmark.

Alzheimer’s and Parkinson’s diseases are two neurodegenerative brain disorders affecting more than 50 million people globally. Early diagnosis and appropriate assessment of disease progression are critical for treatment and improving patients’ health. Currently, the diagnosis of these neurodegenerative diseases is based primarily on mental status exams and neuroimaging scans, which are costly, time-consuming, and sometimes erroneous. Novel, cost-effective, and precise diagnostic tools and techniques are, thus, urgently required, particularly for early detection and prediction. In the recent decade, electromagnetic imaging has evolved as a cost-effective and noninvasive alternative approach for studying brain diseases. These studies focus on wearable and portable devices and imaging algorithms. However, microwave imaging cannot detect minimal changes in the brain at early stages accurately due to its lower resolution. This article investigates machine learning (ML) techniques for the early diagnosis of acute neurological diseases, especially Alzheimer’s disease (AD). A machine-learning-based classification method is proposed. Simulations are performed on realistic numerical brain phantoms using the CST studio suite to get the scattered signals. A novel data augmentation method is proposed to generate synthetic data required for ML algorithms. A deep neural network-based autoencoder extracts features to train various ML algorithms. The classification results are compared with raw data and manual feature extraction. The study shows that the proposed ML-based method could be used to monitor AD at its early stages.

Researchers have explored multiple Wi-Fi features to estimate user locations in indoor environments in the past decade, such as Received Signal Strength Indication (RSSI), Channel State Information (CSI), Time of Arrival (TOA), and Angle of Arrive (AoA). Fine Time Measurement (FTM) is a protocol standardized by IEEE 802.11-2016, which can estimate the distance between the initiator and the station using Wi-Fi Round-Trip Time (RTT). Promoted by Google, such a protocol has been explored in many mobile localization algorithms, which can provide meter-level positioning accuracy between Wi-Fi RTT-enabled smartphones and access points (APs). However, previous studies have shown that the Wi-Fi RTT measurements are sensitive to environmental changes, which leads to significant errors in the localization algorithms. Such an error usually varies according to different environments and settings. Therefore, this paper investigates the error in Wi-Fi RTT distance measurements by setting multiple experiments with different hardware, motion status, and signal path loss conditions. The experiment results show that four categories of errors are found in RTT distance measurements, including hardware-dependent bias, blocker-dependent bias, fluctuations, and outliers. Comparison and analysis are carried out to illustrate the impact of the different errors on Wi-Fi RTT distance.

The smartphone magnetometer has been used in many indoor positioning systems to provide location information, such as orientation, user trajectory construction, and magnetic-field-based fingerprint. However, suffering from magnetic disturbance, magnetometer measurements are vulnerable to interference from metal infrastructures, electrical equipment, and other electronic devices in complex indoor environments. This article extracts and explores the statistical features of the smartphone magnetometer measurements. Extensive experiments in various conditions show that the covariance and the magnitude difference can help detect the magnetic disturbance. Based on this, two unsupervised learning-based methods using Gaussian mixture model and k-means are developed to explore the two features mentioned above in magnetic disturbance detection. Experimental results demonstrate that the two proposed approaches have superior detection accuracy, which is 5%–20% higher than the widely adopted vector selection methods in the literature.

The accuracy of smartphone-based positioning systems using WiFi usually suffers from ranging errors caused by non-line-of-sight (NLOS) conditions. Previous research usually exploits several distribution features from a long time series (hundreds of samples) of WiFi received signal strength (RSS) or WiFi round-trip time (RTT) to achieve a high identification accuracy. However, the long time series or large sample size attributes to high power and time consumption in data collection for both training and testing. This will also undoubtedly be detrimental to user experience as the waiting time for getting enough samples is quite long. Therefore, this paper proposes three new real-time NLOS/LOS identification methods for smartphone-based indoor positioning systems using WiFi RSS and RTT distance measurement (RDM). Based on our extensive analysis of RSS and RDM dispersion features, three machine learning algorithms were chosen and developed to separate the samples for NLOS/LOS conditions. Experiments show that our best method achieves a discrimination accuracy of over 96% with a sample size of 10. Considering the theoretically shortest WiFi ranging interval of 100ms of the RTT-enabled smartphones, our algorithm is able to provide the shortest latency of 1s to get the testing result among all of the state-of-art methods.