Device-Free Occupant Counting Using Ambient RFID and Deep Learning

We present an indoor occupant counting system using ambient radio-frequency identification (RFID) sensors and deep learning models, without requiring on-person tags or movement. We studied the practical settings of both wall and furniture tags. Both received signal strength indicator (RSSI) and phase were calibrated to reduce the interferences from the line-of-sight (LoS) and multi-path components, and the one-hop channel modulation directly caused by the occupants was fed into a convolutional neural network (CNN) for counting. We demonstrated counting accuracies above 90% with 80 tags, and above 85% with 16 - 30 tags in room sizes from 100 to 600 ft(2). Room layouts, RFID tag deployment, and occupants in standing and sitting positions were tested.