Long-range capacitive sensors for indoor person location Mihai Lazarescu . Luciano Lavagno . DET .
Contents Why indoor person localization? System objectives and specifications Capacitive transducers and operating principles Sensor structure and characterization Signal acquisition and processing Localization algorithms Localization experiments Results and conclusions
Why indoor person localization? Improve life quality and safety (home automation, security) Elderly assisted living (detection of routine activity and deviations from it) Sizable market, commercial products
System objectives and specifications Tagless People forget or are reluctant to wear tags Passive Track the person regardless of activities (do not require system interaction) Privacy-aware No pictures, no sound samples Unobtrusive Out of sight, no interference with daily activities
Operation modes of capacitive transducers Transmit and shunt modes require two galvanically- connected plates Difficult and costly installation Load mode senses through one connected plate Simple, low-cost sensor and installation Solutions exist for distances comparable to plate size Too short for target applications
Sensor node structure Copper-clad transducer plate Capacitance-controlled oscillator Frequency proportional to total plate capacitance MCU measures oscillator frequency once per second Sends measurements by radio to server
Sensor characterization (normalized) Three plate sizes: 4x4, 8x8, 16x16 cm Frequency normalized: min. 1, max. 100 Plate-body distance (d): 10 cm to 2 m Capacitance C ~ 1/d^3 noise floor limits sensitivity, mainly for small plates
Sensor characterization (raw)
Server signal acquisition and processing Low-pass 861-taps FIR Eliminates environmental noise High-pass 21-sample median filter Eliminates slow sensor drift
Server signal acquisition and processing 4 sensors in a 3x3 m “room” Different training and localization walking paths Sensor data filtered and normalized Body orientation matters
Localization algorithms (average error) k-Nearest-Neighbors (k-NN) Naïve Bayes (NB) Support Vector Machines (SVM)
Localization algorithms (recall and precision)
Localization algorithms (path reconstruction) 4x4 cm 8x8 cm 16x16 cm k-Nearest- Neighbors (k-NN) Naïve Bayes (NB) Support Vector Machines (SVM)
Next steps Promising results using simple techniques Explore other low-noise high-sensitivity low capacitance variation measurement techniques (ΔC < 100 ppm) Reduce sensor power consumption (energy scavenging?) Reduce overall cost (sensor, deployment) Monitor daily routine in a “real” room Infer typical behaviors Detect and notify deviations