Yoshinori Yaginuma

A robust room-level localization method based on transition probability for indoor environments

Several scene analysis methods [1, 2] such as location fingerprinting have been proposed to execute room-level localization with high accuracy. However, these methods estimate locations wrongly when received signal strength (RSS) observed at a target users current position is similar to RSS in another place. By utilizing the heuristic of passing through a passable boundary point (for example, an entrance door) when moving between rooms, we introduce a technique for evaluating the distance between the current position of the target user and the boundary point by comparing RSS around the boundary point with RSS at the user in order to reduce the number of erroneous judgments. Experiments conducted in office environments confirmed that our proposed technique achieved an accuracy rate of 97.1% in localizing the target user.

Commercial applications on the AP3000 parallel computer

Recently, commercial parallel applications became important in parallel computing with the increase of parallel computer users. In this study, we parallelize two commercial applications, a fulltext search system and a data mining system. In this paper, the implementation of each application and its performance evaluation on the AP3000 parallel computer are shown. A parallel framework, a novel parallelizing approach applied to those applications, is also described.

Eating moment recognition using heart rate responses

There are many studies for recognizing eating moments using wide types of modality (e.g. arm motion). However, they are needed to be improved for both accuracy and robustness for practical use in daily life. In this paper, we propose a novel recognition method using bimodal heart rate responses caused by eating. Our method combines (i) short-term and (ii) long-term features of heart rate changes. The proposed method was evaluated for recognizing eating moment with the free-environment dataset (9 participants, 604 days), and achieved 98.6% accuracy and 56.9% F-score. The proposed features related to ingestion and digestion contribute to robust eating moment recognition.

Continuous Real-Time Measurement Method for Heart Rate Monitoring Using Face Images

This paper investigates fundamental mechanisms of brightness changes in heart rate (HR) measurement from face images through three kinds of experiments; (i) measurement of light reflection from cheek covered with/without copper film, (ii) spectroscopy measurement of reflection light from face and (iii) simultaneous measurement of face images and laser speckle images. The brightness change of the face skin are found to be caused by both the green light absorption variation by the blood volume changes and the light reflection variation by pulsatory face movements. The Real-time Pulse Extraction Method (RPEM), designed to extract the variation of light absorption by removing motion noise, is corroborated for the robustness by comparing the RPEM with the pulse wave of the ear photoplethysmography. The RPEM is also applied to heart rate measurements of seven participants during office work under non-controlled condition in order to evaluate continuous real-time HR monitoring. RMSE = 6.7 bpm is achieved as an average result of seven participants in five days with the 44% of HR measured rate with respect to the number of reference HRs from the electrocardiogram during face is detected. The result indicates that the RPEM method enables HR monitoring in daily life.

Continuous Real-time Heart Rate Monitoring from Face Images

A real-time monitoring method of heart rate (HR) from face images using Real-time Pulse Extraction Method (RPEM) is described and corroborated for the theoretical efficacy by investigating fundamental mechanisms through three kinds of experiments; (i) measurement of light reflection from face covered by copper film, (ii) spectroscopy measurement and (iii) simultaneous measurement of face images and laser speckle images. The investigation indicated the main causes of brightness change are both the green light absorption variation by the blood volume changes and the face surface reflection variation by pulsatory face movements. RPEM removes the motion noise from the green light absorption variation and the effectiveness is ensured by comparing with the pulse wave of the ear photoplethysmography. We also applied RPEM to continuous real-time HR monitoring of seven participants during office work under non-controlled condition, and achieved HR measured rate of 44 % to the number of referential ECG beats while face is detected, with RMSE = 6.7 bpm as an average result of five days.

19.2 cm 3 flexible fetal heart rate sensor for improved quality of pregnancy life

Daily and on-demand monitoring of fetal heart rate (FHR) at home can improve quality of pregnancy life (QoPL), which is sometimes degraded by anxiety about the health conditions of fetus and mother since mothers cannot frequently visit a hospital in order to check the fetal status with expensive ultrasonic measurement. To improve QoPL for mothers, we propose a tiny flexible sensor system that, measures the electrocardiogram (ECG) signal and which is comfortably attached to the mothers abdominal area. The sensor wirelessly transmits the fetal heart rate (FHR) as well as maternal heart rate (MHR) to the mothers smart phone to enable easy monitoring of the fetal status at home. We have developed a patch-type flexible sensor in a small form factor of 72 mm × 38 mm × 7 mm (19.2 cm3) with 13 g, which can feature a secure wireless connection with smart phones via Bluetooth smart, and the FHR and MHR are extracted with an MCU embedded on the sensor to reduce power. Signal processing of the extracted heart rate features analog-domain dual gain paths; the low-gain path extracts the maternal ECG and associated MHR from the original mixed fetal and maternal ECG, and the high-gain path amplifies the fetal ECG up to a level sufficient for reliable peak detection where the clipped maternal ECG is removed from the information via the low-gain path. The simulation results show that 80% of the FHR data is within the specified range (±10%) as compared to the learning data from the PhysioNet database. Since the transmitted data can be reduced from the mixed ECG with a data transmission rate 4.8 kbps to heart rates with transmission rate 8 bps, the percentage of total power consumption taken up by the wireless transmission of data can be reduced from 44% to 9%, allowing for a current consumption of 1.69 mA with a 3.0 V supply and extending the lifetime to 78 hours. Since this sensor is small and bendable, it is suitable for monitoring the maternal and fetal health during daily life.

Estimation of spinal shape profiles in motion using accelerometers

To achieve real-time monitoring of spinal shape profile in daily life we have developed a wireless wearable sensor, small in size, that has a tri-axial accelerometer. By using ten sensors on the surface of the back of the body, a spinal shape profile is represented as a set of angles calculated from the gravitational accelerations measured. In this paper, we propose a new method which removes the effect of motional component by the angle calculation according to a constrained optimization formulation. In our formulation, angles and movable ranges of 19 virtual vertebrae are expressed in constraint conditions. Experiments comparing the spinal shape profiles in standing, lifting, and jumping forward motions demonstrate that our method is comparable to those by optoelectronic motion systems and gives better estimations than those by low-pass filter (LPF). In a jumping forward motion, although the maximum error of LPF is more than 32 cm, our method is less than 9 cm. These results show that our method obtained small errors, even when the LPF shows unrealistic profiles due to the large motional acceleration.