Containerized Computer Vision Applications on Edge Devices

Abstract

The proliferation of IoT devices has led to various computer vision applications, where addressing bandwidth and latency challenges through edge nodes presents significant benefits. However, there are still existing gaps and a need for improvements to optimize IoT applications, especially in the field of computer vision, by overcoming limited resources and enhancing device performance. Addressing these challenges is essential to unlock the full potential of IoT applications in real-world scenarios. This paper evaluates the use of lightweight container technology for computer vision applications which using different algorithms, such as Haar Cascades, HOG and CNN with YOLO algorithm, on edge devices and provides a comprehensive comparison and analysis of different versions of computer vision applications in containers in terms of processing ability, and performance. It focuses on containerizing computer vision applications using Docker to achieve safe execution of multiple applications on these devices without interference and to enable flexibility, efficiency, portability, scalability, and isolation. The study also examines the resource usage, execution time, and receiving time of containerized computer vision applications. The research findings significantly advance our understanding of computer vision processing in IoT and edge computing, thereby opening up new avenues for real-time computing scenarios. These insights have the potential to drive transformative advancements in the field, enabling more efficient and accurate computer vision applications in IoT and paving the way for enhanced real-time decision-making, automation, and intelligent systems.