Abstract—In this paper we propose a novel and systematic approach for dynamic allocation of tasks in a video surveillance system using smart cameras and based on Cloud/Fog architecture. Tracking tasks arrive in the system in a random way and must be assigned to the available devices (cameras, Fog nodes and the Cloud). Our approach guarantees the best solution optimizing power consumption and communication cost over the system. The proposed methods uses an integer programming model and its effectiveness is shown on an application example.