ABSTRACT
While neural networks have brought about impressive advancements in computer vision tasks, these achievements heavily depend on computationally demanding resources, restricting their deployment. The decentralized paradigm of Edge AI computing aims to bring decisional capabilities directly to the edge, facilitating real-time decision-making, streamlined data processing, and reduced dependence on network connectivity. In some cases, it is possible to rely on cloud computing to offload processing tasks, but this can introduce latency issues that affect system responsiveness, security, and efficiency. Instead, searching for optimized neural networks for edge device deployment may lead to a better balance between computational efficiency and accurate analysis, empowering sensors to execute their roles effectively with minimal reliance on external resources. This paper reviews the landscape of deep learning architecture optimization tailored for edge devices. Within this survey, we delve into the state-of-the-art advancements in computer vision techniques optimized for edge computing. The challenges deploying and optimizing computer vision models on edge devices emphasize the importance of efficient computation and resource management while navigating the trade-offs between model performance and hardware constraints.
