Highly Reliable Signal Strength-Based Boundary Crossing Localization in Outdoor Time-Varying Environments

Abstract

Detecting and locating outdoor boundary crossing events is valuable information in curbing drug trafficking, reducing poaching, and protecting high-asset equipment and goods. However, boundary sensing is notoriously challenging, prone to false alarms and missed detections, with serious consequences. Weather events, like rain and wind, make it even more challenging to maintain a low level of missed detections and false alarms. In this paper, we propose and test an automated system of wireless sensors which uses received signal strength (RSS) measurements to localize where a boundary crossing occurs. In addition, we develop new RSS-based statistical models and methods that can quickly be initialized and updated on-line by using link RSS statistics to adapt to time-varying RSS changes due to weather events. These models are implemented in two new classifiers that localize boundary crossings with few missed detections and false alarms. We validate our proposed methods by implementing one of the classifiers in a three month long deployment of a solar-powered, real-time system that captures images of the boundary for ground truth validation. Furthermore, over 75 hours of RSS measurements are collected with an emphasis on collection during weather events, like rain and wind, during which we expect our classifiers to perform the worst. We demonstrate that the proposed classifiers outperform four other baseline classifiers in terms of false alarm probability by 1 to 4 orders of magnitude, and in terms of the misclassification probability by 1 to 2 orders of magnitude.

Unsupervised Learning in RSS-Based DFLT Using an EM Algorithm

“…Estimating the RSS distribution (defined by and ) in two states, when a person is crossing or not crossing the imaginary link line between the transceivers, has been explored in [ 27 , 28 , 29 , 30 , 31 ]. Not only can this information be used to localize people, but also to determine if the monitored area is occupied—a non-trivial task in RF sensing.…”

Never Use Labels: Signal Strength-Based Bayesian Device-Free Localization in Changing Environments

“…As in many model-based DFL methods, MPL adopts the idea that a link is either in an affected state or an unaffected state [13], [15], [16]. However, the novelty in this model is that, given the person's location, the state of the link is not known a priori.…”

“…While fingerprinting can be very accurate in localizing people, the fingerprints must be frequently retrained to stay current in changing environments [12]. Alternative DFL methods like radio tomographic imaging (RTI) [2], Bayesian methods [13], and particle filters [5], [14]- [16] provide more flexibility for DFL because the relationship between measured RSS and a person's location is modeled a priori.…”

Movement Detection Using A Reciprocal Received Signal Strength Model

“…One of the pioneering works in RF sensing demonstrated that people alter the mean and variance of the signal strength [7] and it was shown that anomalies in the measurement statistics can be used to detect the presence of people [8]. Modeling the vacant and occupied states independently improves detection performance [9] and it also enables devicefree localization [10,11,12]. The statistical model also depends on the number of people [3,5], and the linear relationship between the number of people and average signal strength was used in [6] to estimate crowd sizes up to thousands of people.…”