基於改良式HOG之物件追蹤演算法

NCUIR > College of Electrical Engineering & Computer Science > Executive Master of Communication Engineering > Electronic Thesis & Dissertation > Item 987654321/72995

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/72995

Title:	基於改良式HOG之物件追蹤演算法
Authors:	魏國鈞;Wei, Kuo-Chun
Contributors:	通訊工程學系在職專班
Keywords:	物件追蹤;HOG
Date:	2017-01-24
Issue Date:	2017-05-05 17:27:09 (UTC+8)
Publisher:	國立中央大學
Abstract:	摘要物件追蹤在近年來是一個相當熱門話題，特別在電腦視頻領域上，以物件追蹤為基礎用在自動監控系統上更是活躍，其中HOG演算法是目前公認最有效的演算法。實際應用上HOG佔據太多運算，又以Gamma correction最為耗時。本研究本論文設計MHOG(Modified Histogram of Oriented Gradient)，提出用Sobel Edge演算法的概念來延伸來進行取代，修改HOG耗費資源部分，提出一個試用及時系統演算法，稱之為基於改良式HOG之物件追蹤演算法MHOG(Modified Histogram of Oriented Gradient)和HOG相比的優點包括: 1.更能減少運算資源。 2.提升追蹤穩定度在實驗結果上，展示出原始HOG與本研究所改良之差異性。依序是六個測試影像水鳥、汽車(靜止)、計算機、汽車(移動中)、黑猩猩 ;Abstract Object tracking has recently been a widely researched topic, specifically in studies investigating computer-automated surveillance systems in the field of computer vision. Among currently available algorithms , the histogram of oriented gradients (HOG) has been recognized as the most effective design. However, in practice, applying the HOG requires excessive computational resources, particularly for the gamma correction method, which is the most time-consuming method. Therefore, by expanding the concept of Sobel edge detection, the present study proposed a feasible real-time computing system named the modified histogram of oriented gradients (MHOG) to minimize the resource requirements of the conventional HOG. Comparing the MHOG and HOG revealed that the proposed algorithm has lower computational resource requirements and enhances the stability of tracking. In this thesis, the differences between the conventional HOG and MHOD are demonstrated in six experimental images, namely images of a water bird, a stationary automobile, a calculator, a moving automobile, and a chimpanzee.
Appears in Collections:	[Executive Master of Communication Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	481	View/Open

社群 sharing

Loading...