Accurate Driver Detection Exploiting Invariant Characteristics of Smartphone Sensors.

Distracted driving jeopardizes the safety of the driver and others. Numerous solutions have been proposed to prevent distracted driving, but the number of related accidents has not decreased. Such a deficiency comes from fragile system designs where drivers are detected exploiting sensory features from strictly controlled vehicle-riding actions and unreliable driving events. We propose a system called ADDICT (Accurate Driver Detection exploiting Invariant Characteristics of smarTphone sensors), which identifies the driver utilizing the inconsistency between gyroscope and magnetometer dynamics and the interplay between electromagnetic field emissions and engine startup vibrations. These features are invariantly observable regardless of smartphone positions and vehicle-riding actions. To evaluate the feasibility of ADDICT, we conducted extensive experiments with four participants and three different vehicles by varying vehicle-riding scenarios. Our evaluation results demonstrated that ADDICT identifies the driver's smartphone with 89.1% average accuracy for all scenarios and >85% under the extreme scenario, at a marginal cost of battery consumption.

Estimating the Composition of Food Nutrients from Hyperspectral Signals Based on Deep Neural Networks.

There is an increasing demand for acquiring details of food nutrients especially among those who are sensitive to food intakes and weight changes. To meet this need, we propose a new approach based on deep learning that precisely estimates the composition of carbohydrates, proteins, and fats from hyperspectral signals of foods obtained by using low-cost spectrometers. Specifically, we develop a system consisting of multiple deep neural networks for estimating food nutrients followed by detecting and discarding estimation anomalies. Our comprehensive performance evaluation demonstrates that the proposed system can maximize estimation accuracy by automatically identifying wrong estimations. As such, if consolidated with the capability of reinforcement learning, it will likely be positioned as a promising means for personalized healthcare in terms of food safety.

Human-level blood cell counting on lens-free shadow images exploiting deep neural networks.

In point-of-care testing, in-line holographic microscopes paved the way for realizing portable cell counting systems at marginal cost. To maximize their accuracy, it is critically important to reliably count the number of cells even in noisy blood images overcoming various problems due to out-of-focus blurry cells and background brightness variations. However, previous studies could detect cells only on clean images while they failed to accurately distinguish blurry cells from background noises. To address this problem, we present a human-level blood cell counting system by synergistically integrating the methods of normalized cross-correlation (NCC) and a convolutional neural network (CNN). Our comprehensive performance evaluation demonstrates that the proposed system achieves the highest level of accuracy (96.7-98.4%) for any kinds of blood cells on a lens-free shadow image while others suffer from significant accuracy degradations (12.9-38.9%) when detecting blurry cells. Moreover, it outperforms others by up to 36.8% in accurately analyzing noisy blood images and is 24.0-40.8× faster, thus maximizing both accuracy and computational efficiency.

Reliable Identification of Vehicle-Boarding Actions Based on Fuzzy Inference Syste.

Existing smartphone-based solutions to prevent distracted driving suffer from inadequate system designs that only recognize simple and clean vehicle-boarding actions, thereby failing to meet the required level of accuracy in real-life environments. In this paper, exploiting unique sensory features consistently monitored from a broad range of complicated vehicle-boarding actions, we propose a reliable and accurate system based on fuzzy inference to classify the sides of vehicle entrancebyleveragingbuilt-insmartphonesensorsonly. Theresultsofourcomprehensiveevaluation on three vehicle types with four participants demonstrate that the proposed system achieves 91.1%∼94.0% accuracy, outperforming other methods by 26.9%∼38.4% and maintains at least 87.8 %accuracy regardless of smartphone positions and vehicle types.

Smart Gait-Aid Glasses for Parkinson's Disease Patients.

Parkinson's disease (PD) is a chronic progressive disease caused by loss of dopaminergic neurons in the substantia nigra, degenerating the nervous system of a patient over time. Freezing of gait (FOG), which is a form of akinesia, is a symptom of PD. Meanwhile, recent studies show that the gait of PD patients experiencing FOG can be significantly improved by providing the regular visual or auditory patterns for the patients. In this paper, we propose a gait-aid system built upon smart glasses. Our system continuously monitors the gait and so on of a PD patient to detect FOG, and upon detection of FOG it projects visual patterns on the glasses as if the patterns were actually on the floor. Conducting experiments involving ten PD patients, we demonstrate that our system achieves the accuracy of 92.86 % in detecting FOG episodes and that it improves the gait speed and stride length of PD patients by 15.3  âˆ¼  37.2% and 18.7   âˆ¼  31.7%, respectively.

Optimization of a cell counting algorithm for mobile point-of-care testing platforms.

In a point-of-care (POC) setting, it is critically important to reliably count the number of specific cells in a blood sample. Software-based cell counting, which is far faster than manual counting, while much cheaper than hardware-based counting, has emerged as an attractive solution potentially applicable to mobile POC testing. However, the existing software-based algorithm based on the normalized cross-correlation (NCC) method is too time- and, thus, energy-consuming to be deployed for battery-powered mobile POC testing platforms. In this paper, we identify inefficiencies in the NCC-based algorithm and propose two synergistic optimization techniques that can considerably reduce the runtime and, thus, energy consumption of the original algorithm with negligible impact on counting accuracy. We demonstrate that an AndroidTM smart phone running the optimized algorithm consumes 11.5× less runtime than the original algorithm.