Resource Management in Distributed Camera Systems

Sammanfattning: The aim of this work is to investigate different methods to solve the problem of allocating the correct amount of resources (network bandwidth and storage space) to video camera systems. Here we explore the intersection between two research areas: automatic control and game theory. Camera systems are a good example of the emergence of the Internet of Things (IoT) and its impact on our daily lives and the environment. We aim to improve today’s systems, shift from resources over-provisioning to allocate dynamically resources where they are needed the most. We optimize the storage and bandwidth allocation of camera systems to limit the impact on the environment as well as provide the best visual quality attainable with the resource limitations. This thesis is written as a collection of papers. It begins by introducing the problem with today’s camera systems, and continues with background information about resource allocation, automatic control and game theory. The third chapter de- scribes the models of the considered systems, their limitations and challenges. It then continues by providing more background on the automatic control and game theory techniques used in the proposed solutions. Finally, the proposed solutions are provided in five papers.Paper I proposes an approach to estimate the amount of data needed by surveillance cameras given camera and scenario parameters. This model is used for calculating the quasi Worst-Case Transmission Times of videos over a network. Papers II and III apply control concepts to camera network storage and bandwidth assignment. They provide simple, yet elegant solutions to the allocation of these resources in distributed camera systems. Paper IV com- bines pricing theory with control techniques to force the video quality of cam- era systems to converge to a common value based solely on the compression parameter of the provided videos. Paper V uses the VCG auction mechanism to solve the storage space allocation problem in competitive camera systems. It allows for a better system-wide visual quality than a simple split allocation given the limited system knowledge, trust and resource constraints.