The Future of Camera Networks

Abstract

Camera networks become smart when they can interpret video data on board, in order to carry out tasks as a collective, such as target tracking and (re-)identification of objects of interest. Unlike today's deployments, which are mainly restricted to lab settings and highly controlled high-value applications, future smart camera networks will be messy and unpredictable. They will operate on a vast scale, drawing on mobile resources connected in networks structured in complex and changing ways. They will comprise heterogeneous and decentralised aggregations of visual sensors, which will come together in temporary alliances, in unforeseen and rapidly unfolding scenarios. The potential to include and harness citizen-contributed mobile streaming, body-worn video, and robotmounted cameras, alongside more traditional fixed or PTZ cameras, and supported by other non-visual sensors, leads to a number of difficult and important challenges. In this position paper, we discuss a variety of potential uses for such complex smart camera networks, and some of the challenges that arise when staying smart in the presence of such complexity. We present a general discussion on the challenges of heterogeneity, coordination, self-reconfigurability, mobility, and collaboration in camera networks. KEYWORDSfuture camera networks, surveillance, mobile smart cameras, body worn cameras, humans in the loop, distributed control ACM Reference format:

Edge Intelligence: Challenges and Opportunities of Near-Sensor Machine Learning Applications

“…2 by increasing the number of filters from one to six, we observe a small increase on the number of BFLOPs from 0.002(1) BFLOPs to 0.009 (6) BFLOPs. By increasing the size of each filter from one to six we observe an increase from 0.002(1) BFLOPs to 0.057 (6) BFLOPs. This shows a significant impact on the performance.…”

“…Additionally, this is limiting for applications processing private data, such as wearable biomedical devices, or wearable cameras [6]. Processing private user data at the edge could protect user privacy and avoid exposing the data to security threats by sending them to the cloud.…”

A Collective Adaptive Approach to Decentralised k-Coverage in Multi-robot Systems

“…In [32], Esterle and Lewis rely on purely distributed approaches. They enable the individual robots to learn about their local environment, including other robots and analyse the potential of the topological neighbourhood of interaction.…”