Wireless camera sensors networks are an emerging sensing technology that could enable new applications in security, transportation, and healthcare. Tracking and identifying moving objects is a fundamental visual surveillance task and methods that respect the energy and bandwidth constraints of a wireless sensor network are needed. This paper proposes a realtime algorithm for tracking people and acquiring canonical frontal face images in this setting. Clusters of sensors collaborate to track individuals and groups of people moving in an indoor environment. The sensors are tasked to retrieve the best frontal face image for each tracked individual according to a score based on face size and orientation. The method exploits information about the target�s trajectory to retrieve an approximate best frontal face image in an energy efficient manner. Frontal face images acquired by this algorithm are suitable for standard face recognition algorithms and would be valuable for identity management. To evaluate this approach on real data, a prototype surveillance system called FaceNet was developed. FaceNet displays a compact summary of human activity by overlaying a floorplan diagram with the 2D trajectories and a face image for each track. A simple benchmark of FaceNet shows the amount of data transmitted from the sensors can be reduced by 97 percent compared to a naive centralized streaming architecture and has the potential to significantly reduce the energy used by wireless nodes for this type of surveillance task.