This work presents a scalable, distributed architecture for processing spatiotemporal queries in the context of camera networks based on a graph structure. With the ever-increasing presence of cameras and the emergence of camera-networks, e.g., in the context of campus security, it becomes increasingly important to provide a robust and scalable architecture to store and retrieve detected events. In this work a distributed graph processing engine will be presented which is well suited for read and write tasks in the environment of spatiotemporal image-similarity based workloads. The key ideas presented in this work are the architecture of a scalable graph processing system well-suited for processing spatio-temporal queries and the design of a distributed and robust vertex-partitioning strategy for the graph which is being defined by the spatiotemporal attributes of the stored events. The work will show multiple lightweight heuristics for partitioning the graph among the nodes participating in the system, focusing on load-balancing between workers and high edge-locality for vertices. The system and the partitioning strategies will be evaluated, showing that the system scales with the number of workers and the problem size and is able to answer proportionally more queries per second. It will also be shown that the lightweight heuristics for partitioning the graph produce a relatively good balancing of the vertices on the worker-nodes and can be executed in an online-fashion, resulting in similar performance when compared to a traditional hash-partitioning while providing far superior edge-locality.