English  |  正體中文  |  简体中文  |  Items with full text/Total items : 26988/38789
Visitors : 2358264      Online Users : 36
RC Version 4.0 © Powered By DSPACE, MIT. Enhanced by NTU Library IR team.
Scope Adv. Search
LoginUploadHelpAboutAdminister

Please use this identifier to cite or link to this item: http://ntour.ntou.edu.tw:8080/ir/handle/987654321/49406

Title: 使用深度學習資料探勘與對稱SURF軌跡之動作分析技術
Action Classification Using Deep Learning, Data Mining and Symmetrical SURF Based Trajectories
Authors: Salah Alghyaline
林時樂
Contributors: NTOU:Department of Computer Science and Engineering
國立臺灣海洋大學:資訊工程學系
Keywords: 人類行為識別;關聯規則挖掘;密集軌跡;頻繁項集;深度學習;異常場景變化檢測
human action recognition;association rule mining,;dense trajectories;frequent itemsets;SURF;deep learning;abnormal scene changing detection
Date: 2017
Issue Date: 2018-08-22T06:57:25Z
Abstract: 在本論文中,提出了六種算法:(1)在 數據庫中挖掘頻繁項集,(2)基於挖掘的動作識別,(3)基於對稱的動作識別,(4)基於相似性的動作識別,(5)Symmelet基於視頻動作識別的深度軌跡,以及(6)使用補丁 (bags of patches)和蜘蛛網圖 (spider-web map)來檢測異常場景變化。與頻繁項目挖掘中的可用方法相比,所提出的挖掘頻繁模式技術能夠顯著減少計算時間和記憶體使用量,並得到相同的挖掘結果,所提出的方法遍歷FP樹(FP-tree) 而不需構建任何條件FP樹(FP-tree)、FP增長(FP-growth)。而與已知的密集軌跡Dense Trajectories(DT), trajectory-pooled deep-convolutional descriptor (TDD),和改進的密集軌跡Improved Dense Trajectories(IDT)方法相比,四種提出的動作識別方法提高了最具挑戰性的數據集中的動作識別準確度,而採用的方法是通過發現視頻特徵之間的語義關係,來提高動作分類精度,像使用關聯規則挖掘和視頻特徵之間的對稱性。最後,使用提出的異常場景變化檢測方法的實驗結果表明,該方法在預定義環境中有效且準確地發現異常物。
In this dissertation, six algorithms are proposed: (1) mining frequent itemsets in transactional databases, (2) mining-based action recognition, (3) symmetry-based action recognition, (4) similarity-based action recognition, (5) Symmelet-based deep trajectories for video action recognition, and (6) using bags of patches and spider-web map for detecting abnormal scene changes. Compared with the available approaches in frequent itemsets mining, the proposed mining frequent pattern technique is able to reduce the computation and the memory time significantly and gives the same mining result, the proposed method traverses the FP-tree without building any conditional FP-tree compared with FP-growth. Whereas the four proposed action recognition approaches increase the action recognition accuracy in the most challenging datasets compared with the known dense trajectories (DT), improved dense trajectories (IDT), and trajectory-pooled deep-convolutional descriptor (TDD) approaches. The proposed methods enhance the action classification accuracy by discovering semantic relationships between video features like using association rule mining, and the symmetry property between video features. Finally, the experimental results based on the proposed method of detecting abnormal scene changes show that the method is efficient and accurate to discover abnormal objects from a mobile camera mounted on a robot.
URI: http://ethesys.lib.ntou.edu.tw/cgi-bin/gs32/gsweb.cgi?o=dstdcdr&s=G0020157003.id
http://ntour.ntou.edu.tw:8080/ir/handle/987654321/49406
Appears in Collections:[資訊工程學系] 博碩士論文

Files in This Item:

File Description SizeFormat
index.html0KbHTML11View/Open


All items in NTOUR are protected by copyright, with all rights reserved.

 


著作權政策宣告: 本網站之內容為國立臺灣海洋大學所收錄之機構典藏,無償提供學術研究與公眾教育等公益性使用,請合理使用本網站之內容,以尊重著作權人之權益。
網站維護: 海大圖資處 圖書系統組
DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library IR team Copyright ©   - Feedback