本文選題：特征提取 + 特征表達��； 參考：《安徽工程大學(xué)》2017年碩士論文

【摘要】：隨著當今社會數(shù)字化和信息化程度的不斷提高,視覺信息越來越多以數(shù)字圖像的形式存在于人們?nèi)粘５纳罴吧a(chǎn)中。作為圖像處理和機器人視覺的重要組成部分,特征提取技術(shù)可以對視覺信息進行高效的處理,獲得人們需要的信息,給生活和工業(yè)生產(chǎn)帶來便利。它在工業(yè)機器人視覺技術(shù)、衛(wèi)星遙感技術(shù)、信息檢索技術(shù)及圖像處理等方面應(yīng)用非常廣泛。作為工業(yè)機器人的眼睛,機器人視覺技術(shù)是工業(yè)機器人必不可少的,而特征提取技術(shù)是工業(yè)機器人視覺領(lǐng)域最重要的技術(shù)之一。本文以工業(yè)機器人對物體的識別和分類為應(yīng)用背景,結(jié)合來自文本處理領(lǐng)域的BOW(Bag of Words,詞袋模型)模型,對工業(yè)機器人視覺特征提取技術(shù)展開研究。以SIFT(Scale Invariant Feature Transformation,尺度不變特征變換)特征提取算法為主要切入點,針對SIFT算法自身存在的缺陷對其做出改進。同時,結(jié)合SIFT特征將文本領(lǐng)域中的BOW模型進行重新設(shè)計,形成一種新的特征表達模型,使得SIFT特征提取算法能更好應(yīng)用于工業(yè)機器人中。本文的主要研究內(nèi)容及創(chuàng)新性如下:(1)本課題針對目前工業(yè)機器人對于復(fù)雜背景環(huán)境下物體識別和判斷能力不足的問題對特征提取算法展開研究。通過大量的閱讀文獻和試驗,確定以對復(fù)雜背景下圖像特征提取相對較好的SIFT算法為主要入手點展開研究。(2)針對SIFT特征提取算法對邊緣及非線性變化光照處理能力不足的缺點,結(jié)合Laplacian邊緣算子改善邊緣特征提取效果不好的問題;同時,提出一種反正切歸一化法代替原有的歸一化方法,來改善非線性變化光照條件對特征提取干擾大的問題。(3)經(jīng)過大量的研究發(fā)現(xiàn)SIFT特征在機器人物體識別領(lǐng)域應(yīng)用較多是在物體的匹配上,因為SIFT特征不能被現(xiàn)有的分類器直接分類。針對這一點,本文利用文本分類領(lǐng)域的BOW模型結(jié)合PCA(Principal Component Analysis,主成分分析)技術(shù),提出一種新的特征表達模型。將SIFT提取的視覺特征重新進行表達,再輸入分類器進行分類,從而實現(xiàn)對物體的識別和判斷。(4)針對現(xiàn)存工業(yè)機器人安全防護技術(shù)的不足,結(jié)合本文研究成果提出一種基于機器人視覺技術(shù)的工業(yè)機器人安全防護技術(shù)的方案,并通過實驗驗證了該方案的可行性。以上的研究及創(chuàng)新內(nèi)容最后以實驗的方式進行驗證。實驗結(jié)果表明,本文對算法的改進是成功的,設(shè)計的特征表達模型具有可靠性,本文的研究成果具有實際應(yīng)用價值。
[Abstract]:With the development of digitization and informatization, more and more visual information exists in people's daily life and production in the form of digital images. As an important part of image processing and robot vision, feature extraction technology can efficiently process visual information, obtain the information that people need, and bring convenience to life and industrial production. It is widely used in industrial robot vision, satellite remote sensing, information retrieval and image processing. As the eyes of industrial robot, robot vision technology is indispensable to industrial robot, and feature extraction is one of the most important technologies in the field of industrial robot vision. In this paper, based on the recognition and classification of objects by industrial robots, combined with the Bow bag of Wordsmodel (word bag model) model from the field of text processing, the visual feature extraction technology of industrial robots is studied. Based on the sift / scale variant feature transformation (scale invariant feature transformation) algorithm, this paper improves the sift algorithm in view of its defects. At the same time, a new feature representation model is formed by redesigning the Bow model in the text domain with sift features, which makes the sift feature extraction algorithm better applied to industrial robots. The main contents and innovations of this paper are as follows: (1) in this paper, we study the feature extraction algorithm for the problem that the current industrial robot has insufficient ability to recognize and judge objects in complex background. Through reading a lot of literature and experiments, it is determined that the sift algorithm, which is relatively good for image feature extraction in complex background, is the main starting point of the research. (2) aiming at the shortcoming of sift feature extraction algorithm for edge and nonlinear illumination processing, it is pointed out that sift feature extraction algorithm has insufficient ability to deal with edge and nonlinear variation illumination. Combining with Laplacian edge operator to improve the effect of edge feature extraction is a problem, at the same time, a new method is proposed to replace the original normalization method. After a lot of research, it is found that sift feature is widely used in the field of robot object recognition in object matching. Because sift features cannot be directly classified by existing classifiers. In order to solve this problem, a new feature representation model is proposed in this paper, which is based on the Bow model in the field of text classification and PCA-Principal component Analysis (PCA) technology. The visual features extracted by sift are re-expressed, and then input into the classifier to classify the objects, so as to realize the recognition and judgment of objects. (4) aiming at the shortcomings of the existing safety protection technology of industrial robots, Based on the research results of this paper, a scheme of industrial robot safety protection based on robot vision technology is proposed, and the feasibility of the scheme is verified by experiments. The above research and innovation content is verified by experiment. The experimental results show that the improved algorithm is successful and the designed feature representation model is reliable. The research results in this paper have practical application value.
【學(xué)位授予單位】：安徽工程大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP391.41;TP242.2

【參考文獻】

相關(guān)期刊論文 前10條

1 徐健鋒;許園;許元辰;張遠健;劉清;;基于語義理解和機器學(xué)習(xí)的混合的中文文本情感分類算法框架[J];計算機科學(xué);2015年06期

2 Fei-wei QIN;Lu-ye LI;Shu-ming GAO;Xiao-ling YANG;Xiang CHEN;;A deep learning approach to the classification of 3D CAD models[J];Journal of Zhejiang University-Science C(Computers & Electronics);2014年02期

3 王民;劉偉光;;基于改進SIFT特征的雙目圖像匹配算法[J];計算機工程與應(yīng)用;2013年02期

4 陳凱;肖國強;潘珍;李正浩;;單尺度詞袋模型圖像分類方法[J];計算機應(yīng)用研究;2011年10期

5 曲永宇;劉清;郭建明;周生輝;;基于HOG和顏色特征的行人檢測[J];武漢理工大學(xué)學(xué)報;2011年04期

6 趙欽君;趙東標;韋虎;;Harris-SIFT算法及其在雙目立體視覺中的應(yīng)用[J];電子科技大學(xué)學(xué)報;2010年04期

7 何友松;吳煒;陳默;楊曉敏;羅代升;;基于Bag of Features算法的車輛圖像識別研究[J];電視技術(shù);2009年12期

8 劉婧;秦華標;;基于灰度直方圖的幾何特征人眼定位方法[J];微計算機信息;2009年09期

9 張五一;趙強松;王東云;;機器視覺的現(xiàn)狀及發(fā)展趨勢[J];中原工學(xué)院學(xué)報;2008年01期

10 蔡鶴皋;機器人技術(shù)的發(fā)展與在制造業(yè)中的應(yīng)用[J];機械制造與自動化;2004年01期

相關(guān)博士學(xué)位論文 前1條

1 楊釗;面向圖像分類和識別的視覺特征表達與學(xué)習(xí)的研究[D];華南理工大學(xué);2014年

相關(guān)碩士學(xué)位論文 前3條

1 汪松;基于SIFT算法的圖像匹配方法研究[D];西安電子科技大學(xué);2013年

2 高錦;基于SVM的圖像分類[D];西北大學(xué);2010年

3 宰小濤;基于SIFT特征描述子的立體匹配算法研究[D];上海交通大學(xué);2007年

，

本文編號：1993319