An Efficient Pipeline for Distant Person Detection and Identification in 4K Video using GPUs

Govardhan Mattela, Manmohan Tripathi, Chandrajit Pal, Rampelli Sai Dhiraj, A. Acharyya
{"title":"An Efficient Pipeline for Distant Person Detection and Identification in 4K Video using GPUs","authors":"Govardhan Mattela, Manmohan Tripathi, Chandrajit Pal, Rampelli Sai Dhiraj, A. Acharyya","doi":"10.1109/COMSNETS48256.2020.9027465","DOIUrl":null,"url":null,"abstract":"The gradual advent of machine learning has been assisting to shift the field of computer vision from statistical methods to deep neural networks. These networks should be able to process high resolution video streams coming from the HD camera sources in real time. However, due to the fixed network size and to maintain the processing speed, high resolution frames need to be resized and down-sampled before feeding into the networks resulting in loss of feature information, hampering recognition accuracy. This motivated us to propose a methodology which focuses on creating and processing the active region of interests in the foreground image through an active region generator (ARG) module, eliminating the need to traverse the entire frame and down-sample the resolution before feeding it to the neural network. This resulted in saving 25x more image feature information, whilst maintaining a given person detection accuracy of 92 % mAP for longer distance up to 30~35 metre executing in real time w.r.t it's classical counterpart based on singleshot detector model. Besides, our proposed pipeline architecture utilizing multi-core TESLA GPU increases the execution throughput by a factor of 3X verified in NVIDIA DGX system.","PeriodicalId":265871,"journal":{"name":"2020 International Conference on COMmunication Systems & NETworkS (COMSNETS)","volume":"122 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2020-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2020 International Conference on COMmunication Systems & NETworkS (COMSNETS)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/COMSNETS48256.2020.9027465","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 1

Abstract

The gradual advent of machine learning has been assisting to shift the field of computer vision from statistical methods to deep neural networks. These networks should be able to process high resolution video streams coming from the HD camera sources in real time. However, due to the fixed network size and to maintain the processing speed, high resolution frames need to be resized and down-sampled before feeding into the networks resulting in loss of feature information, hampering recognition accuracy. This motivated us to propose a methodology which focuses on creating and processing the active region of interests in the foreground image through an active region generator (ARG) module, eliminating the need to traverse the entire frame and down-sample the resolution before feeding it to the neural network. This resulted in saving 25x more image feature information, whilst maintaining a given person detection accuracy of 92 % mAP for longer distance up to 30~35 metre executing in real time w.r.t it's classical counterpart based on singleshot detector model. Besides, our proposed pipeline architecture utilizing multi-core TESLA GPU increases the execution throughput by a factor of 3X verified in NVIDIA DGX system.
基于gpu的4K视频中远距离人物检测与识别的高效流水线
机器学习的逐渐出现有助于将计算机视觉领域从统计方法转移到深度神经网络。这些网络应该能够实时处理来自高清摄像机源的高分辨率视频流。然而,由于网络规模固定,为了保持处理速度,高分辨率帧在输入网络之前需要调整大小和下采样,导致特征信息丢失,影响识别精度。这促使我们提出了一种方法,该方法侧重于通过有源区域生成器(ARG)模块在前景图像中创建和处理感兴趣的有源区域,从而消除了在将其输入神经网络之前遍历整个帧并对分辨率进行下采样的需要。这导致节省了25倍的图像特征信息,同时保持了92% mAP的给定人员检测精度,最远可达30~35米,实时执行,而不是基于单发探测器模型的经典对应。此外,我们提出的利用多核TESLA GPU的流水线架构将执行吞吐量提高了3倍,在NVIDIA DGX系统中得到了验证。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
自引率
0.00%
发文量
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:604180095
Book学术官方微信