Off-campus UNL users: To download campus access dissertations, please use the following link to log into our proxy server with your NU ID and password. When you are done browsing please remember to return to this page and log out.
Non-UNL users: Please talk to your librarian about requesting this dissertation through interlibrary loan.
Bayes’ Network and Smart Sensors – Occupancy Detection
Abstract
This dissertation presents a Bayesian analysis for determining residential occupancy using inexpensive commercially available passive infrared (PIR) motion detectors, compared against two other detectors that were used to establish ground-truth. One of the ground-truth detectors was a GPS signal from a smartphone, the second was a Bluetooth key fob. Data were gathered from four residential locations, and then analyzed to determine occupancy. The occupancy data collected from the PIR sensors were compared against ground-truth to verify the results of the PIR sensor events that were collected every minute for a week. The Bayesian training data that was used to determine the prior probability used a four-week time period collected once a minute. Having established the correspondence between ground-truth and the PIR sensor events, the PIR data were then used to build Bayesian network conditional tables. Once the conditional tables were constructed, the Bayesian network results could be compiled and then compared against the ground-truth data. One analysis compared the ground-truth data against the performance of individual PIR sensors and showed that there was a low correlation between the PIR motion and occupancy. Further analyses compared the ground-truth data against the performance of various groupings of PIR sensors within each residence and showed that there was a little less correlation than the individual PIR sensors method. When Bayesian modeling was applied using historical PIR sensor data, results demonstrated an improvement in occupancy detection over the individual and grouped PIR sensor methods that were evaluated. The historical sensor data (using PIR sensor signal pulses) was successfully applied to the network, with an average of .025 ϕ correlation improvement. The historical presence data (using ground-truth data) were then applied to the same network. This step improved the ϕ correlation between the PIR sensors and ground-truth by an average of .40 over the four locations. These findings show that applying Bayesian modeling improves the accuracy of occupancy detection required for safety and efficiency, which will permit occupants to live in their homes longer.
Subject Area
Engineering
Recommended Citation
Tryon, Donald Levi, "Bayes’ Network and Smart Sensors – Occupancy Detection" (2020). ETD collection for University of Nebraska-Lincoln. AAI28258302.
https://digitalcommons.unl.edu/dissertations/AAI28258302