Researchers at Colorado State University have developed a novel secured methodology to maintain indoor localization accuracy, even in the presence of AP attacks. The proposed methodology consists of 1) Extrapolation and creation of the Secure AP Attack Resilient (SAAR) database; and 2) Training secure-CNN models (S-CNN) that are resilient to malicious AP based attacks.
Indoor localization is a much needed tool for various situations, including navigating students to classrooms (non-critical) or medical staff and equipment to patients in real time (critical).
Indoor localization is an emerging application domain for the navigation and tracking of people and assets. Ubiquitously available Wi-Fi signals have enabled low-cost fingerprinting-based localization solutions. Further, the rapid growth in mobile hardware capability now allows high-accuracy deep learning-based frameworks to be executed locally on mobile devices in an energy-efficient manner. However, existing deep learning-based in-door localization solutions are vulnerable to access point (AP) attacks.
Malicious third parties can exploit the vulnerabilities of unsecured indoor localization components (e.g., Wi-Fi Access Points or WAPs) to produce incorrect localization information. This may lead to some inconvenience in non-critical scenarios (e.g., a student arrives at the wrong classroom), but can lead to dire consequences in more critical situations (e.g., medical staff are unable to locate vital equipment or medicine needed fora patient in an emergency; or emergency response personnel are misdirected, causing a loss of lives).
Hackers, however, may not be the only cause of vulnerabilities. Some vulnerabilities may arise from temporary and unintentional environmental changes that may alter the accuracy of indoor localization techniques. This tainted information from intentional or unintentional sources can lead to even more egregious real-time delays and errors. Therefore, similar to outdoor navigation systems, establishing secure and reliable indoor localization and navigation systems holds an uncontested importance in this domain.
The secure convolutional neural network (CNN) based indoor localization solution extrapolates the offline fingerprint database, to obtain a larger number of samples per reference point. This creates a larger fingerprint database to train the network, producing better results. Then noise is deliberately introduced, which helps improve the robustness of the network against minor deviations or noise in images that can be induced during malicious attacks. This improved training procedure produced a model that consistently reduces the vulnerability of the proposed localization framework. Results demonstrated a localization accuracy 25 times better than an unsecured CNN-based localization framework, even in the presence of threats from malicious attackers.
Available for Licensing
Sudeep Pasricha
Saideep Tiku
Jessy McGowan
Jessy.McGowan@colostate.edu
970-491-7100