基於 YOLO 演算法在多種場景下的危險物品即時偵測

以作者查詢圖書館館藏

、以作者查詢臺灣博碩士

、以作者查詢全國書目

、勘誤回報

、線上人數：61

、訪客IP：3.138.172.222

姓名

呂政穆(Cheng-Mu Lu) 查詢紙本館藏

畢業系所

通訊工程學系

論文名稱

基於 YOLO 演算法在多種場景下的危險物品即時偵測
(Real-Time Detection of Danger Objects in Various Scenes Based on the YOLO Algorithm)

相關論文

★ UHF頻段RFID彈藥管理系統之設計、實作與評估	★ 移動物偵測與追蹤之IP Camera系統
★ SDN自適應性自動化網路安全之研究	★ Wi-Fi Direct Service 應用於IoT
★ 射頻前端電路應用於載波聚合長期演進技術	★ 3C無線充電裝置運用在車載系統所產生之EMI輻射
★ 基於LoRa技術的物聯網前端防盜警示感測裝置實作與評估	★ DOCSIS 3.1 效能研究與下行通道干擾阻隔之設計
★ 藍芽無線光學投影翻譯筆	★ 手持裝置應用於MIMO ( 8x8 ) Wi-Fi系統之設計
★ 基於無伺服器運算之智慧農業雲端系統設計與研究	★ 嵌入式系統實現電梯物聯網
★ 在802.11 Ad-Hoc網路中基於速率考量之路由協定設計	★ 合作博弈與灰色模糊方法改善無線網路之性能
★ 採用拍賣策略之動態分散式方法於減少叢集小型基地台間干擾之研究	★ 在LTE-A下聚合未授權頻譜及動態分配資源以優化系統效能

檔案

[Endnote RIS 格式]

[Bibtex 格式]

[相關文章]

[文章引用]

[完整記錄]

[館藏目錄]

至系統瀏覽論文 (2026-7-10以後開放)

摘要(中)

隨著網路世界高速發展，多種社群平台走進大眾的視野之中，像是
Youtube 和 TikTok，在物聯網下比起以往的文字貼文，影音的影響更是大
幅的增加資訊的普及性，無論是各個學術領域的知識亦或是新聞資訊的熱
門影片，搭配短影片的出現更是加速傳播的速度，導致許多未成年人可能
會持續認知這些偏頗的資訊且未能被及時糾正，進而導向錯誤的價值觀。
本論文提出一個使用 YOLO 透過增量式訓練搭配不同優化器組合選
擇，針對危險物品訓練出高準確率的偵測模型，透過對螢幕畫面進行擷取
的方式，使其可以針對多場景下的影像進行即時偵測並辨識，像是針對
Twitch 串流平台上和 Youtube 影音平台上的影像，透過辨識出危險物品
後，對其進行圖像擷取並加以標註圖像訊息，透過 Line Notify 進行通報使
用者。
本系統能夠縮減各種硬體上的限制，不同於以往透過監視器進行人為
審查的部分，能大幅降低所需的人力成本和時間成本，並且高準確率能減
少人為篩檢的錯誤概率，在實際操作上也展現出其即時性和高準確率，這
種偵測模式不單單受限於家長監護管理，也適用於任何需要偵測的場景。

摘要(英)

With the rapid development of the internet, various social platforms like
YouTube and TikTok have come into the public eye. Under the Internet of
Things, the impact of videos has significantly increased the dissemination of
information. This applies to academic fields of knowledge and popular news.
The advent of short videos has further accelerated the spread of information,
leading to many minors potentially continuously absorbing biased information
without timely correction, which can lead to distorted values.
This paper proposes a high-accuracy detection model for dangerous objects
trained using YOLO through incremental training with different Optimizer
combinations. By capturing screen images, this system can perform real-time
detection and recognition of images across multiple scenarios, such as on
various social platforms. Upon recognizing dangerous objects, the system
captures and annotates the images, and notifies users through Line Notify.
This system reduces various hardware limitations from traditional manual
monitoring through surveillance cameras, significantly reducing required human
and time costs. The high accuracy reduces the probability of errors. In practical
application, it demonstrates its real-time capability and high accuracy. This
detection model is not only limited to parental supervision but is also applicable
to any scenario requiring detection.

關鍵字(中)

★ YOLO
★ 物聯網
★ 增量式訓練
★ 優化器組合
★ 即時影像偵測

關鍵字(英)

★ YOLO
★ IoT
★ Incremental Learning
★ Optimizer combinations
★ Real-time detection

論文目次

目錄
摘要 I
Abstract II
誌謝 III
目錄 IV
圖目錄 VII
表目錄 IX
第一章序論 1
1.1 前言 1
1.2 研究動機 2
1.3 研究方向 3
1.4 論文架構 4
第二章相關研究背景 5
2.1 影像辨識步驟及介紹 5
2.1.1 物件辨識 7
2.1.2 人臉辨識 9
2.1.3 YOLO 11
2.2 機器學習 15
2.2.1 資料強化 16
2.2.2 卷積神經網路CNN 17
2.3 OBS & Xsplit VCam 18
2.4 AWS 19
2.4.1 Amazon S3 19
2.5 Line 20
2.5.1 Line Notify 20
第三章系統架構與驗證 21
3.1 系統架構 21
3.2 系統訓練端 23
3.2.1 收集危險物品圖片 24
3.2.2 危險物品數據庫 25
3.2.3 資料強化 26
3.2.4 設定訓練參數 28
3.2.5 訓練危險物品模型 29
3.2.6 AWS同步模型 29
3.2.7 模組效能評估指標-混淆矩陣 30
3.3 系統偵測辨識端 32
3.3.1 即時影像擷取輸入 33
3.3.2 影像偵測並辨識 34
3.3.3 目標物件處理並發送Line Notify 35
第四章模擬與分析 37
4.1 模擬設定 37
4.1.1 訓練參數調整及設定 37
4.1.2 目標擷取參數設定 39
4.2 模擬結果 40
4.2.1 YOLOv5l之於YOLOv7之性能比較 40
4.2.2 系統整體執行效能 45
第五章結論與未來研究方向 49
參考文獻 50

參考文獻

[1] J. Redmon, S. Divvala, R. Girshick and A. Farhadi, "You Only Look Once： Unified, Real-Time Object Detection," 2016 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）, Las Vegas, NV, USA, 2016, pp. 779-788, doi： 10.1109/CVPR.2016.91
[2] L. Li, X. Mu, S. Li and H. Peng, "A Review of Face Recognition Technology," in IEEE Access, vol. 8, pp. 139110-139120, 2020, doi： 10.1109/ACCESS.2020.3011028.
[3] L. Jiao et al., "A Survey of Deep Learning-Based Object Detection," in IEEE Access, vol. 7, pp. 128837-128868, 2019, doi： 10.1109/ACCESS.2019.2939201.
[4] K. He, G. Gkioxari, P. Dollár and R. Girshick, "Mask R-CNN," 2017 IEEE International Conference on Computer Vision （ICCV）, Venice, Italy, 2017, pp. 2980-2988, doi： 10.1109/ICCV.2017.322.
[5] 網路資料on line resources：one stage和 two stage流程圖。取自https：//github.com/bourdakos1/Custom-Object-Detection https：//github.com/bourdakos1/Custom-Object-Detection
[6] Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C., & Berg, A. C. （2015）. SSD： Single Shot MultiBox Detector. ArXiv. https：//doi.org/10.1007/978-3-319-46448-0_2
[7] 網路資料on line resources：單一物件處理跟多重物件處理流程圖取自。https：//cs231n.stanford.edu/schedule.html
[8] Insaf, Adjabi & Ouahabi, A. & Benzaoui, Amir & taleb-ahmed, Abdelmalik. （2020）. Past, Present, and Future of Face Recognition： A Review.
[9] 網路資料on line resources：人臉辨識流程圖張凱喬，人臉辨識-基本流程/測試標準取自。https://weilihmen.medium.com/%E4%BA%BA%E8%87%89%E8%BE%A8%E8%AD%98-%E5%9F%BA%E6%9C%AC%E6%B5%81%E7%A8%8B-%E6%B8%AC%E8%A9%A6%E6%A8%99%E6%BA%96-8d4d7c66e8ff
[10] 網路資料on line resources：NMS流程圖 Tommy Huang，機器/深度學習：物件偵測 Non-Maximum Suppression（NMS）。取自https：//chih-sheng-huang821.medium.com/%E6%A9%9F%E5%99%A8-%E6%B7%B1%E5%BA%A6%E5%AD%B8%E7%BF%92-%E7%89%A9%E4%BB%B6%E5%81%B5%E6%B8%AC-non-maximum-suppression-nms-aa70c45adffa
[11] Bochkovskiy, A., Wang, C., & Liao, H. （2020）. YOLOv4： Optimal Speed and Accuracy of Object Detection. ArXiv. /abs/2004.10934
[12] 網路資料on line resources：Pytorch。取自https://pytorch.org/
[13] Wang, C., Bochkovskiy, A., & Liao, H. （2022）. YOLOv7： Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. ArXiv. /abs/2207.02696
[14] L. Chen, P. Chen and Z. Lin, "Artificial Intelligence in Education： A Review," in IEEE Access, vol. 8, pp. 75264-75278, 2020, doi： 10.1109/ACCESS.2020.2988510.
[15] K. He, X. Zhang, S. Ren and J. Sun, "Deep Residual Learning for Image Recognition," 2016 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）, Las Vegas, NV, USA, 2016, pp. 770-778, doi： 10.1109/CVPR.2016.90.
[16] Perez, L., & Wang, J. （2017）. The Effectiveness of Data Augmentation in Image Classification using Deep Learning. ArXiv. /abs/1712.04621
[17] Xue Ying 2019 J. Phys,” An Overview of Overfitting and its Solutions”,Conf. Ser. 1168 022022
[18] 網路資料on line resources：OBS取自。https://obsproject.com/
[19] A. Aloman, A. I. Ispas, P. Ciotirnae, R. Sanchez-Iborra and M. D. Cano, "Performance Evaluation of Video Streaming Using MPEG DASH, RTSP, and RTMP in Mobile Networks," 2015 8th IFIP Wireless and Mobile Networking Conference （WMNC）, Munich, Germany, 2015, pp. 144-151, doi： 10.1109/WMNC.2015.12.

[20] 網路資料on line resources：XSplit VCam取自。https://www.xsplit.com/?gad_source=1&gclid=CjwKCAjw1emzBhB8EiwAHwZZxZaWGgJ2uakYCEnK09wg92UFH4IPIRSlTrHWZ2bi5XzNDZcmMRRO4BoCocMQAvD_BwE&pp=stripe_affiliate
[21] 網路資料on line resource：AWS取自。https：//aws.amazon.com/tw/
[22] 網路資料on line resource：Amazon S3取自。https：//aws.amazon.com/tw/s3/
[23] 網路資料on line resource：Line取自。https：//zh.wikipedia.org/zh-tw/LINE
[24] 網路資料on line resource：Line Notify取自。https：//notify-bot.line.me/zh_TW/
[25] 網路資料on line resource：Unsplash取自。https：//unsplash.com/
[26] Keskar, N. S., & Socher, R. （2017）. Improving Generalization Performance by Switching from Adam to SGD. ArXiv. /abs/1712.07628
[27] A. O. Ramon and L. Barba Guaman, "Detection of weapons using Efficient Net and Yolo v3," 2021 IEEE Latin American Conference on Computational Intelligence (LA-CCI), Temuco, Chile, 2021, pp. 1-6
[28] A. Warsi, M. Abdullah, M. N. Husen, M. Yahya, S. Khan and N. Jawaid, "Gun Detection System Using Yolov3," 2019 IEEE International Conference on Smart Instrumentation, Measurement and Application (ICSIMA), Kuala Lumpur, Malaysia, 2019, pp. 1-4

指導教授

吳中實(Jung-Shyr Wu)

審核日期

2024-7-15

推文