2018 Fiscal Year Annual Research Report
Activity recognition based on a small scale sparse sensor network
| Project/Area Number |
16K00334
|
|---|
| Research Institution | The University of Aizu |
|---|
Principal Investigator |
趙 強福 会津大学, コンピュータ理工学部, 教授 (90260421)
|
|---|
| Project Period (FY) |
2016-04-01 – 2019-03-31
|
| Keywords | Sensor array / Smart home / Monitoring system / Privacy preserving / Senior care / Machine learning |
|---|
| Outline of Annual Research Achievements |
In the first year (2016), we conducted experiments using a sensor array containing 10 ZigBee motion sensors. Experimental results show that the sensor array can provide information for a machine learner (e.g. multilayer neural network) to recognize a limited number fixed locations. However, since the data rate of the sensors is very slow (10 data per second), it was difficult to collect many data for various locations. In the experiments we also found that the ZigBee sensors were not stable, and data loss has occured frenquently in the experiment.
To solve the problems, in 2017, we designed a new sensor array containing 3x5 motion sensors. Each sensor is controlled by an Aruduino compatible board, and USB is used for both data communication and power supply. Compared with the ZigBee-based sensor, the data rate is about 100 times faster, there is no data loss, and we can conduct experiments without worrying about the statues of the batteries. Using this sensor array, we studied location and activity-strength recognition/estimation, and obtained very good results.
To make the sensor array more practically useful, we designed a compact sensor array module and filed a patent. As an example, a 6-sensor module can participate a 4x4 square meter space into 19 sub-regions, and the sensor outputs can be installed more easily on the ceiling. We have fabricated a prototype using 3D printers. The size (diameter) of the module is 10cm. We are now trying to improve the "productivity" as well as the security of the module, and commercialize the module as soon as possible.
|
