Human, Activity and Environment Sensing

In recent years, the development of wearable sensors has made it possible to easily collect a variety of biological data. We are studying how to collect these biometric data in various environments, such as sports gyms and universities, and use them to improve the estimation of deep body temperature using wearable sensors, to detect signs of heat stroke, and to estimate stress levels.

In recent years, various wearable devices have become popular and are expected to be used for applications such as monitoring the elderly and healthcare. To solve this problem, we are trying to improve the sensing performance by coordinating multiple environmental sensors driven by environmental power generation, while eliminating the need for recharging. 

It has been found that the power consumption for sensor coordination can be significantly reduced by using backscatter communication, which has been attracting attention recently. We propose a new method of sensor coordination based on the backcatter communication method. In this research, we are cooperating with these environment-generating sensors and implementing a classifier that has been trained by machine learning in a microcontroller to estimate the movement status such as "walking" or "riding a bus or train.

We are also working on distributed execution of deep learning in wireless sensor networks. In wireless sensor networks, if the microcontrollers attached to the sensors become more sophisticated and power-efficient, it will be possible to realize an autonomous intelligent sensor network in which tasks such as learning, anomaly detection, and judgment, which were previously performed in the cloud, are offloaded to the sensor network and can be efficiently performed in a location close to where the data is generated. Intelligent sensor networks can be realized.

In recent years, the big data in various sports has been utilized to improve the performance of players and to create advanced tactical plans.

For example, we are working on a play-scene recognition using soccer tracking data, and player position estimation during wheelchair basketball using 9-axis sensors (3-axis acceleration, 3-axis angular velocity, and 3-axis geomagnetism). In the position estimation, the displacement and direction of each player are estimated for each unit of time and the trajectory of each player's movement is derived from the data obtained from the sensors attached to the bottom of the wheelchair and the axles of both wheels and the wheelchair shape measured in advance. 

We are studying localization and behavior estimation using Wi-Fi radio waves.

For example, we use Wi-Fi radio waves to identify people and objects by spatially capturing the reflection strength. Focusing on the fact that the reflection of radio waves has different characteristics depending on the material of the object, we attach tags to people and objects that are made of different materials with different radio wave reflection characteristics and are arranged separately for each object. The receiver captures the intensity of the reflected waves in two dimensions and identifies the object by reading the arrangement of the tags. In this way, even when there are multiple objects, each can be identified.

We are also developing tags that use a technology called Backscatter, which is an ultra-low-power communication technology that switches between two states, like the reflection of a mirror, to influence the surrounding radio waves and vice versa. By attaching tags to people and objects, each tag generates a different variation of radio waves, thereby separating the effects of multiple objects. Furthermore, by acquiring a time-series of the variations of multiple reflected waves, we can recognize the surrounding situation, such as "person A and person B are having a conversation" or "person C is watching TV," based on the positional relationship of specific people or objects. We are also exploring new approaches to wireless sensing. For example, by switching the switch on and off in response to a person's movement, we believe that it will be possible to monitor the elderly in medical and nursing care facilities without the need to change the batteries of sensor devices.