The Problem of Dynamic Spatial Sampling and Geofence Surveillance

TL;DR

This paper proposes adaptive radius estimators for geofence surveillance that balance law enforcement needs with individual privacy by considering local human activity density.

Contribution

It introduces optimal radius estimators based on empirical human activity distributions to improve privacy-aware geofence surveillance.

Findings

Estimators adapt surveillance perimeters to local human activity density.

Using these estimators can reduce privacy risks in geofence surveillance.

The approach offers a method for law enforcement to balance surveillance effectiveness and privacy.

Abstract

Geofencing surveillance poses a dynamic spatial sampling problem. Law enforcement must establish geofence perimeters to identify a relevant suspect. This requires identifying a sampling region around a surveillance site and counting the number of intersecting individuals as proxied by geolocation tags. Law enforcement commonly constructs sampling regions with fixed distance intervals or fixed polygon boundaries. This generates privacy concerns as considerations for constructing these perimeters do not factor in the local density of human activity, such as pedestrian flows or traffic patterns. This increases the risk of selective expansion where agencies attempt to extend their data collection beyond what a warrant previously approved. This paper attempts to balance law enforcement's needs for surveillance with individual level privacy by proposing a set of optimal radius estimators. These plug-in estimators use the empirical distribution of human activity patterns to estimate an optimal radius. Given a surveillance site and set of point densities, the optimal radius generates surveillance perimeters that adapt to local conditions. We discuss the implications of applying this estimator to policing surveillance efforts and how law enforcement can use algorithms to better protect the privacy of its citizens.