Robust identification of human activities in buildings using multiple sensor system

As a first step to realizing "Biofication of Living Space" that needs spatial recognition for accommodation and evolution, a new human identification system was developed. The system can identify human activities using information from multiple sensors. Acceleration sensors, pyroelectric infrared sensors and microphone were adopted in this study. And these sensors' information was fused and handled together. Therefore, the system recognized 10 persons with up to 91.3 percent accuracy by 21 dimensional feature quantities. And also, it was confirmed that by combination of multiple sensors' information, the false discrimination rate was reduced by about half compared to single sensor's information. Simultaneous identification of multi persons was performed. Using the extracted feature quantities, the discrimination rate was calculated in 3 settings of learn data. When learn data was extracted data of measurement that tested 4 persons walked alone, the system recognizes 4 persons with up to 80 percent accuracy. Moreover, even in situation that lacked data from some sensors human identification can be achieved with more than 80 percent accuracy.