Participatory sensing applications rely on individuals to share local and personal data with others to produce aggregated models and knowledge. In this setting, privacy is an important consideration, and lack of privacy could discourage widespread adoption of many exciting applications. We present a privacy-preserving participatory sensing scheme for multidimensional data which uses negative surveys. Multidimensional data, such as vectors of attributes that include location and environment fields, are challenging for privacy protection and are common in participatory sensing applications. When reporting data in a negative survey, an individual participant randomly selects a value from the set complement of the sensed data value, once for each dimension, and returns the negative values to a central collection server. Using algorithms described in this paper, the server can reconstruct the probability density functions of the original distributions of sensed values, without knowing the participants' actual data. Our algorithms avoid computationally expensive encryption and key management schemes, conserving energy. We study trade-offs between accuracy and privacy, and their relationships to the number of dimensions, categories, and participants. We introduce dimensional adjustment, a method that reduces the magnification of error associated with earlier work. Two simulation scenarios illustrate how the approach can protect the privacy of a participant's multidimensional data while allowing useful aggregate information to be collected.

Sensor networks involving human participants will require privacy protection before wide deployment is feasible. This paper proposes and evaluates a set of protocols that enable anonymous data collection in a sensor network. Sensor nodes, instead of transmitting their actual data, transmit a sample of the data complement to a basestation. The basestation then uses the negative samples to reconstruct a histogram of the original sensor readings. These protocols, collectively defined as a negative survey, are computationally simple and do not increase communication overhead. Thus, the negative survey can be implemented efficiently on existing sensor network platforms. We analyze the accuracy of the negative survey under a variety of conditions and define a range of parameter values for which it is practical. We also describe an example traffic monitoring application that uses the negative survey to classify traffic behavior. We demonstrate that for reasonable traffic scenarios, the system accurately classifies traffic behavior without revealing private information.

No abstract available.