Vinaya Gohokar and Vijay Gohokar
Edge intelligence, object detection, Internet of Things (IoT), artificial intelligence, Jetson Nano, IoMT
Implementation of artificial intelligence-based algorithms on resource-constrained devices at the edge of the network is a challenging task. This paper reviews architecture, models, and requirements to implement edge intelligence-based application. A comparative study of edge devices, Raspberry pi (Rpi) with neural compute stick (NCS), and Jetson Nano (Nano) for object detection and recognition using deep learning models is presented. Incisive observations regarding impact of optimisation frameworks, libraries, and selection of the right device depending on the application are discussed. Inference time, energy use, and cost effectiveness are compared. The results obtained using Tensor-RT on Jetson Nano have proved promising for IoMT applications. Mobile Net v2 model achieves best performance.
Important Links:
Go Back
