Abstract-Distributed aggregation algorithms have traditionally been applied to environments with no or rather low rates of node churn. The proliferation of mobile devices in recent years introduces high mobility and node churn to these environments, thus imposing a new dimension on the problem of distributed aggregation in terms of scalability and convergence speed. To address this, we present DiVote, a distributed voting protocol for mobile device-to-device communication. We investigate a particular use case, in which pedestrians equipped with mobile phones roam around in an urban area and participate in a distributed yes/no poll, which has both spatial and temporal relevance to the community. Each node casts a vote and collects votes from other participants in the system whenever in communication range; votes are immediately integrated into a local estimate. The objective of DiVote is to produce a precise mapping of the local estimate to the anticipated global voting result while preserving node privacy. Since mobile devices may have limited resources allocated for mobile sensing activities, DiVote utilizes D-GAP compression. We evaluate the proposed protocol via extensive trace-driven simulations of realistic pedestrian behavior, and demonstrate that it scales well with the number of nodes in the system. Furthermore, in densely populated areas the local estimate of participants does not deviate by more than 3 % from the global result. Finally, in certain scenarios the achievable compression rate of DiVote is at least 19 % for realistic vote distributions.