本文選題：車載全景系統(tǒng) + 圖像配準(zhǔn)　； 參考：《重慶理工大學(xué)》2017年碩士論文

【摘要】：隨著汽車數(shù)量的迅速增加,由于駕駛員視覺盲區(qū)引發(fā)的安全駕駛問題日益突出、交通事故不斷攀升。目前,車載全景系統(tǒng)被認(rèn)為是緩解該問題的有效手段之一,已成為智能交通和圖像處理領(lǐng)域近年來的研究熱點(diǎn),從而吸引了眾多學(xué)者的關(guān)注。而圖像配準(zhǔn)是決定車載全景系統(tǒng)是否可行的關(guān)鍵環(huán)節(jié)。因此,本文針對(duì)車載全景系統(tǒng)中快速圖像配準(zhǔn)方法開展了深入研究,主要研究?jī)?nèi)容如下:1、針對(duì)傳統(tǒng)尺度不變特征變換(SIFT)算法在特征提取與特征描述時(shí)計(jì)算量大、實(shí)時(shí)性差的問題,提出一種基于區(qū)域分塊的SIFT快速配準(zhǔn)方法。首先,將匹配圖像和待匹配圖像分割成若干均勻的子圖,通過計(jì)算每個(gè)子圖的信息熵值與設(shè)定閾值比較來判斷局部子圖的特征類型;然后對(duì)篩選出來的子圖進(jìn)行SIFT特征提取和對(duì)特征向量降維處理,生成PCA-SIFT描述子;最后對(duì)兩幅圖像進(jìn)行配準(zhǔn)和剔除誤匹配點(diǎn)對(duì)。結(jié)果表明,在保證配準(zhǔn)精度在90%以上的基礎(chǔ)上,配準(zhǔn)時(shí)間減少了15%-25%。2、針對(duì)傳統(tǒng)的Harris角點(diǎn)檢測(cè)算法,手動(dòng)輸入單個(gè)閾值可能出現(xiàn)角點(diǎn)聚簇、偽角點(diǎn)等現(xiàn)象。提出了一種自適應(yīng)閾值的Harris角點(diǎn)檢測(cè)的圖像配準(zhǔn)方法,首先將圖像分割成3?3個(gè)無重疊子圖,根據(jù)每個(gè)子圖的對(duì)比度的大小,來設(shè)置每個(gè)子圖的閾值。然后采用NCC算法對(duì)檢測(cè)出的角點(diǎn)進(jìn)行粗匹配;最后采用RANSAC算法將粗匹配中誤匹配點(diǎn)對(duì)進(jìn)行剔除。實(shí)驗(yàn)表明,該算法使檢測(cè)的角點(diǎn)分布比較均勻,并在圖像配準(zhǔn)中有效地增加圖像匹配點(diǎn)對(duì)數(shù),能有效提高圖像配準(zhǔn)的準(zhǔn)確度。3、為了驗(yàn)證本文的圖像配準(zhǔn)算法在車載全景系統(tǒng)中應(yīng)用效果,采用Directshow技術(shù)和兩個(gè)USB攝像頭搭載了系統(tǒng)實(shí)驗(yàn)平臺(tái),分別采用改進(jìn)的SIFT算法、改進(jìn)的Harris角點(diǎn)檢測(cè)算法對(duì)采集的圖像幀進(jìn)行提取特征點(diǎn)和圖像配準(zhǔn)的效果對(duì)比。實(shí)驗(yàn)結(jié)果表明:采用改進(jìn)的SIFT配準(zhǔn)算法,融合后圖像效果良好,并能有效減少圖像拼接時(shí)間。總之,本文研究的圖像快速配準(zhǔn)算法,可提高圖像配準(zhǔn)效率20%左右,有效減少了圖像配準(zhǔn)環(huán)節(jié)的時(shí)間消耗,為研制實(shí)用的車載圖像全景系統(tǒng)奠定技術(shù)基礎(chǔ)。
[Abstract]:With the rapid increase of the number of cars, the problem of safe driving caused by drivers' visual blind area is becoming more and more serious, and the traffic accidents are rising. At present, vehicle panoramic system is regarded as one of the effective means to alleviate this problem, and it has become the research hotspot in the field of intelligent transportation and image processing in recent years, which has attracted the attention of many scholars. Image registration is the key to determine the feasibility of vehicle panoramic system. Therefore, in this paper, the fast image registration method in vehicle panoramic system is deeply studied. The main research contents are as follows: 1. For the traditional scale-invariant feature transformation (SIFT) algorithm, the computation of feature extraction and feature description is heavy. A fast SIFT registration method based on regional block is proposed. Firstly, the matching image and the image to be matched are divided into several uniform subgraphs, and the feature types of the local subgraphs are determined by calculating the information entropy value of each subgraph and comparing the information entropy value with the set threshold value. Then SIFT feature extraction and feature vector dimensionality reduction are used to generate the PCA-SIFT descriptor. Finally, the two images are registered and the mismatched point pairs are eliminated. The results show that the registration time is reduced by 15% to 25% on the basis of the registration accuracy is over 90%. For the traditional Harris corner detection algorithm, the single threshold of manual input may appear such phenomena as corner clustering and pseudo corner. An image registration method for adaptive threshold Harris corner detection is proposed. Firstly, the image is divided into 3? 3 non overlapping subgraphs. The threshold of each subgraph is set according to the contrast of each subgraph. Then the NCC algorithm is used to coarse match the detected corner, and the RANSAC algorithm is used to eliminate the mismatched points in rough matching. Experiments show that the algorithm makes the corner distribution more uniform, and effectively increases the logarithm of image matching points in image registration. It can effectively improve the accuracy of image registration. In order to verify the application effect of the image registration algorithm in vehicle panoramic system, Directshow technology and two USB cameras are used to carry the system experimental platform, and the improved SIFT algorithm is adopted respectively. The improved Harris corner detection algorithm is used to extract feature points and image registration. The experimental results show that the improved SIFT registration algorithm is effective and can effectively reduce the time of image stitching. In a word, the fast image registration algorithm studied in this paper can improve the efficiency of image registration by about 20%, effectively reduce the time consumption of image registration, and lay a technical foundation for the development of a practical panoramic image system.
【學(xué)位授予單位】：重慶理工大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：U463.6;TP391.41

【參考文獻(xiàn)】

中國(guó)期刊全文數(shù)據(jù)庫 前10條

1 王麗芳;趙雅楠;秦品樂;高媛;;基于ORB和GroupSAC復(fù)雜場(chǎng)景視頻圖像的快速角點(diǎn)檢測(cè)[J];科學(xué)技術(shù)與工程;2017年02期

2 汪軍;梁鳳梅;;基于P樣條和局部互信息的非剛性醫(yī)學(xué)圖像配準(zhǔn)[J];計(jì)算機(jī)應(yīng)用研究;2017年08期

3 李玉峰;李廣澤;谷紹湖;龍科慧;;基于區(qū)域分塊與尺度不變特征變換的圖像拼接算法[J];光學(xué)精密工程;2016年05期

4 趙小強(qiáng);岳宗達(dá);;一種面向圖像拼接的快速匹配算法[J];南京理工大學(xué)學(xué)報(bào);2016年02期

5 李欽;游雄;李科;張彥喜;;PCA-SIFT特征匹配算法研究[J];測(cè)繪工程;2016年04期

6 張東;余朝剛;;基于特征點(diǎn)的圖像拼接方法[J];計(jì)算機(jī)系統(tǒng)應(yīng)用;2016年03期

7 張勇;王志鋒;馬文;;基于改進(jìn)SIFT特征點(diǎn)匹配的圖像拼接算法研究[J];微電子學(xué)與計(jì)算機(jī);2016年03期

8 杜往澤;宋執(zhí)環(huán);閆文博;吳樂剛;;單攝像頭旋轉(zhuǎn)監(jiān)控下的快速圖像拼接[J];中國(guó)圖象圖形學(xué)報(bào);2016年02期

9 劉坤;呂曉琪;谷宇;于荷峰;任國(guó)印;張明;;快速數(shù)字影像重建的2維/3維醫(yī)學(xué)圖像配準(zhǔn)[J];中國(guó)圖象圖形學(xué)報(bào);2016年01期

10 吳一全;王凱;;基于SUSAN算子和角點(diǎn)判別因子的目標(biāo)邊緣檢測(cè)[J];中國(guó)科學(xué)院大學(xué)學(xué)報(bào);2016年01期

中國(guó)碩士學(xué)位論文全文數(shù)據(jù)庫 前4條

1 胡玉;車載圖像系統(tǒng)中的全景影像的校正與拼接[D];吉林大學(xué);2016年

2 仲明;基于特征點(diǎn)精確配準(zhǔn)的圖像拼接技術(shù)的研究[D];華東師范大學(xué);2015年

3 王則浪;基于DM8148的車載全景拼接算法及視頻記錄軟件設(shè)計(jì)與實(shí)現(xiàn)[D];浙江工業(yè)大學(xué);2014年

4 林蔚;基于互信息和蟻群算法的多分辨率二維—三維醫(yī)學(xué)圖像配準(zhǔn)的研究[D];廣西大學(xué);2013年

，

本文編號(hào)：1944688