Triangulation for k-anonymity in location trajectory data

Abstract

An apparatus for providing anonymity in geographic data for probe devices in a geographic region for a location-based service includes at least a database and a triangulation calculator. The database is configured to store trajectory data based on sequences of sensor measurements of the probe devices. The triangulation calculator is configured to calculate triangles from the trajectory data. Each of the triangles is defined by vertices. The triangulation calculator is configured to calculate a similarity score for the vertices based on trajectory data associated with adjacent vertices. At least one vertex is modified in response to the similarity score.

Description

FIELD[0001]The following disclosure relates to anonymity for probe data for location-based services.BACKGROUND[0002]The Global Positioning System (GPS) or another global navigation satellite system (GNSS) provides location information to a receiving device anywhere on Earth as long as the device has a substantial line of sight without significant obstruction to three or four satellites of the system. Location-based services control features of an application based on location information from a GNSS or another source.[0003]The increasing trend of smart phones and wide spread integration of GPS devices in vehicles lead to availability of large pool of user location data including stay-points, checkins and mobility traces. When such mobility data is aggregated in a centralized manner, it makes new applications such as traffic analysis and prediction possible. The aggregated and sharing of mobility traces data is called trajectory data publishing.[0004]Trajectory data publishing is cen...

AI Technical Summary

Benefits of technology

The patent describes a method for anonymizing geographic data collected from probe devices in a specific region for a location-based service. The method involves creating triangles from the trajectory data of the probe devices based on their sensor measurements. The triangles are then assigned geographic locations based on their centrality score, which is calculated by comparing the trajectory data of adjacent vertices. By doing this, the method allows for more accurate and reliable location tracking while protecting the privacy of the collected data. The technical effect of this patent is to improve the accuracy and accuracy of location-based services while protecting the privacy of the collected data.

Problems solved by technology

Though the usefulness of such publishing is unquestionable, the trade-off being made is privacy of contributing users and control over data.

When it is possible to trace back the contributing user it might lead to unintended consequences that harms the safety and security of contributors.

A data breach by any of these applications may result in the identity and habits of the users being compromised.

In practice, this does little to mitigate the risks because an adversary may break the protection and identify the user with external information.

A repeatedly observed trip from a residential place to a business place may suggest the home location and office location; frequent visits to a hospital may suggest health issues.

In another aspect, strong privacy preserving techniques involving data perturbation, lead to noise data that is published impacting the utility of shared data.

K-anonymization and clustering methods have a strong connection, as the underlying problem is determining an optimal grouping that best represents the structure of the data.

Releasing or publishing an unchanged version of the sequence of spatiotemporal coordinates may violate usual privacy (e.g., reveal the identities of one or more persons, devices, or entities).

Achieving such anonymity optimally is a challenging problem and is NP-hard (e.g., the time for solving the problem of anonymity is bounded by a polynomial expression in the size of the input and the solution is quickly solvable).

This property makes anonymizing mechanisms challenging for location data.

The closest assignment results in uneven attribution in terms of membership, so the number of representatives may be lower than K. Because these points may violate anonymity property, the representatives are removed so that the members that are assigned to them are reassigned to a better representative in the successive step, introducing distortion as a side effect.

The characterization and clustering phase of the first embodiment is not tuned optimally to underlying data and is sensitive to user parameters.

Also trajectories with high frequency sampling might lead to oversampling and over-representation of those instances.

In such a case, it is not possible to optimize the tree further and hence considered as a stopping criterion.

Additionally, the illustrations are merely representational and may not be drawn to scale.

Images