國立虎尾科技大學 |

Towards Achieving End-to-End Edge AI for Computer Vision via System Integration of Real Time Video Feeds With AI Models Aided by Image Enhancement.

Record Type:	Language materials, manuscript : Monograph/item
Title/Author:	Towards Achieving End-to-End Edge AI for Computer Vision via System Integration of Real Time Video Feeds With AI Models Aided by Image Enhancement./
Author:	Sanderson, Jonathan.
Description:	1 online resource (80 pages)
Notes:	Source: Masters Abstracts International, Volume: 85-06.
Contained By:	Masters Abstracts International85-06.
Subject:	Computer engineering. -
Online resource:	click for full text (PQDT)
ISBN:	9798381168068

Towards Achieving End-to-End Edge AI for Computer Vision via System Integration of Real Time Video Feeds With AI Models Aided by Image Enhancement.
Sanderson, Jonathan.

Towards Achieving End-to-End Edge AI for Computer Vision via System Integration of Real Time Video Feeds With AI Models Aided by Image Enhancement. - 1 online resource (80 pages)

Source: Masters Abstracts International, Volume: 85-06.

Thesis (M.S.)--Tennessee Technological University, 2023.

Includes bibliographical references

Deployment of Convolutional Neural Networks for Computer Vision (CV) tasks is an integral part of many modern applications in edge computing (e.g. self-driving cars). Xilinx provides a development platform, that combines FPGA fabric and ARM processors to compute complex CV tasks in real-time. There has been literature for real-time High Definition Multimedia Interface (HDMI) input, deep processing unit (DPU), and HDMI output. However, the literature is deficient in providing a systematic methodology for end-to-end software/hardware integration of HDMI input/output, DPU, with image enhancement. This work begins by integrating image enhancement and DPU using the Gstreamer pipeline, along with creating a Graphical User Interface (GUI) to make the pipeline more intuitive. Next, the effect of using two common image enhancement algorithms, Histogram Equalization (HE) and Retinex, on CNNs for countering adverse weather conditions is explored by simulating dark, over-exposure, hazy, and dark & rainy weather conditions. The change in performance is measured across several CNNs, Resnet, Googlenet, YOLO, and a Vision Transformer (ViT). Lastly, this work provides an end-to-end methodology and implementation for system integration of HDMI input/output, DPU, and image enhancement using FPGA fabric. For this, a HE IP has been developed and tested with Resnet50 and YOLOv3 to improve accuracy when inferencing dark images. A frame rate of over ten frames per second is seen, along with a 30% accuracy improvement when HE is used to enhance video frames, versus not using HE. To promote open research we plan to provide source code used in this work.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798381168068Subjects--Topical Terms:

569006
Computer engineering.
Subjects--Index Terms:

Convolutional neural networksIndex Terms--Genre/Form:

554714
Electronic books.

Towards Achieving End-to-End Edge AI for Computer Vision via System Integration of Real Time Video Feeds With AI Models Aided by Image Enhancement.
LDR:03074ntm a22003857 4500 001 1148330
005 20240924101855.5
006 m o d
007 cr bn ---uuuuu
008 250605s2023 xx obm 000 0 eng d
020 $a 9798381168068
035 $a (MiAaPQ)AAI30694117
035 $a AAI30694117
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Sanderson, Jonathan. $3 1474274
245 1 0 $a Towards Achieving End-to-End Edge AI for Computer Vision via System Integration of Real Time Video Feeds With AI Models Aided by Image Enhancement.
264 0 $c 2023
300 $a 1 online resource (80 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Masters Abstracts International, Volume: 85-06.
500 $a Advisor: Hasan, Syed Rafay.
502 $a Thesis (M.S.)--Tennessee Technological University, 2023.
504 $a Includes bibliographical references
520 $a Deployment of Convolutional Neural Networks for Computer Vision (CV) tasks is an integral part of many modern applications in edge computing (e.g. self-driving cars). Xilinx provides a development platform, that combines FPGA fabric and ARM processors to compute complex CV tasks in real-time. There has been literature for real-time High Definition Multimedia Interface (HDMI) input, deep processing unit (DPU), and HDMI output. However, the literature is deficient in providing a systematic methodology for end-to-end software/hardware integration of HDMI input/output, DPU, with image enhancement. This work begins by integrating image enhancement and DPU using the Gstreamer pipeline, along with creating a Graphical User Interface (GUI) to make the pipeline more intuitive. Next, the effect of using two common image enhancement algorithms, Histogram Equalization (HE) and Retinex, on CNNs for countering adverse weather conditions is explored by simulating dark, over-exposure, hazy, and dark & rainy weather conditions. The change in performance is measured across several CNNs, Resnet, Googlenet, YOLO, and a Vision Transformer (ViT). Lastly, this work provides an end-to-end methodology and implementation for system integration of HDMI input/output, DPU, and image enhancement using FPGA fabric. For this, a HE IP has been developed and tested with Resnet50 and YOLOv3 to improve accuracy when inferencing dark images. A frame rate of over ten frames per second is seen, along with a 30% accuracy improvement when HE is used to enhance video frames, versus not using HE. To promote open research we plan to provide source code used in this work.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Computer engineering. $3 569006
650 4 $a Electrical engineering. $3 596380
653 $a Convolutional neural networks
653 $a Edge artificial intelligence
653 $a GStreamer
653 $a Image enhancement
655 7 $a Electronic books. $2 local $3 554714
690 $a 0464
690 $a 0544
690 $a 0800
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a Tennessee Technological University. $b Electrical Engineering. $3 1184626
773 0 $t Masters Abstracts International $g 85-06.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30694117 $z click for full text (PQDT)