Robust Distributed Dictionary Learning for In-Network Image Compression

Camera networks are resource-constrained distributed systems that communicate over (wireless) networks to make decisions collaboratively. For surveillance applications, these camera nodes take decisions about an object of interest within incoming videos by coordinating with neighboring nodes, which is a costly process in terms of both time and energy. Datacompression methods can bring significant energy savings in camera nodes while transmitting or storing data in the network. Signal representation using sparse approximations and overcomplete dictionaries have received considerable attention in recent years and have been shown to outperform traditional compression methods. However, distributed dictionary learning itself relies on consensus-building algorithms, which involve communicating with neighboring nodes until convergence is achieved. To this end, we design a novel protocol to enable energy-efficient and robust dictionary learning in distributed camera networks by leveraging the spatial correlation of collected multimedia data. We employ low-computational-complexity metrics to quantify the correlation across cameras nodes. We also present a feasibility study of the parameters of the network that impact the performance of distributed dictionary learning and consensus process in terms of accuracy of the algorithm and energy consumed by the camera nodes. The performance of the proposed approach is validated through extensive simulations using a network simulator and public datasets as well as via real-world experiments on a testbed of Raspberry Pi nodes.