Deep learning inference on edge device: Traffic violation detection using OpenVino

• Main Author: Chiew, Jing Cheng
• Format: Final Year Project / Dissertation / Thesis
• Published: 2023
• Subjects: Q Science (General); T Technology (General)
• Online Access: http://eprints.utar.edu.my/5516/1/fyp_CS_2023_CJC.pdf; http://eprints.utar.edu.my/5516/

Deep learning technologies are becoming increasingly popular in recent years. Numerous industries, including healthcare, entertainment, automation systems, natural language processing, and others, are impacted by it. It advances global technology to a new level. Deep learning techniques are now widely employed as a result, especially on edge devices that perform IoT tasks. It is because we no longer need people to assist us in our work, we instead choose to deploy an edge device with a deep learning model. To run the code effectively without being bothered by the slow processing times, those deep learning approaches demand for a lot of processing power, which requires strong computer hardware. This project interprets and demonstrates how OpenVino (Open Visual Inference and Neural) toolkits assist in improving performance and enable us to run a demanding deep learning model on an Intel’s computer system that most regular people have. The OpenVino’s inference engine is designed to speed up the inference of deep learning models under IR format that is provided by OpenVino. This project will explain whether the OpenVino toolkit does indeed offer a shorter inference time and eventually how much performance can be delivered. Before the project ends, a traffic violation detection application that combined with several deep learning pre-trained models will be configured and deployed in an intel-powered edge device (Intel UP board). It aims to determine whether running an OpenVino-optimized deep learning framework application on an edge device with a low-power processor can surprisingly produce a respectable result.