本文關(guān)鍵詞： 局部特征 特征點(diǎn)檢測 協(xié)方差描述子 特征點(diǎn)匹配 三維物體識(shí)別　出處：《中北大學(xué)》2017年碩士論文　論文類型：學(xué)位論文

【摘要】：三維物體識(shí)別在計(jì)算機(jī)視覺中是一項(xiàng)十分重要的基礎(chǔ)性研究,它是對遠(yuǎn)程遙感、生物醫(yī)療、機(jī)器人在內(nèi)的等眾多領(lǐng)域進(jìn)行研究的前提和基礎(chǔ),因此三維物體識(shí)別有著廣泛的應(yīng)用前景。3D物體識(shí)別方法大致可以分為兩類:基于全局特征的識(shí)別方法和基于局部特征的識(shí)別方法。由于基于全局特征的識(shí)別方法忽略了物體的一些局部信息,所以在一些雜亂的、存在遮擋的場景中,不能很好的識(shí)別,而基于局部特征的三維物體識(shí)別技術(shù)有更大的優(yōu)勢�；诰植刻卣鞯娜S物體識(shí)別方法大體上可以分為三個(gè)階段:3D特征點(diǎn)的檢測,特征點(diǎn)的描述和局部曲面的匹配。本文圍繞這三個(gè)核心階段,對基于局部特征的三維物體識(shí)別方法進(jìn)行了深入的研究,主要做了以下幾個(gè)方面的工作。(1)針對物體的尺度不變性特征以及傳統(tǒng)算法對噪聲敏感等問題,本文提出了一種基于散亂點(diǎn)云的多尺度特征點(diǎn)提取算法,通過改變局部鄰域的大小來構(gòu)造尺度空間進(jìn)行多尺度分析,在不同的尺度下通過對局部鄰域的協(xié)方差分析來計(jì)算曲面變化值,找到具有尺度不變性的特征點(diǎn)。同時(shí)還引入了基于形狀索引值的點(diǎn)簽名方法,增強(qiáng)了對噪聲的魯棒性。(2)針對傳統(tǒng)描述子維數(shù)過大,匹配時(shí)間較長等問題。本文提出了一種幾何協(xié)方差描述子,利用特征點(diǎn)與鄰域點(diǎn)間法向量夾角、距離等幾何特征構(gòu)造協(xié)方差矩陣來描述特征點(diǎn)。實(shí)驗(yàn)表明,此描述子不僅具有較強(qiáng)的描述能力,而且具有旋轉(zhuǎn)平移不變性,對噪聲不敏感,對點(diǎn)云采樣密度也具有較強(qiáng)的魯棒性。(3)在曲面匹配過程中,針對特征點(diǎn)最近鄰匹配后還存有一定的誤匹配對,本文將典型相關(guān)分析引入到特征點(diǎn)的誤匹配剔除上來,最終得到了較好的匹配效果。
[Abstract]:3D object recognition is a very important basic research in computer vision. It is the premise and foundation of remote sensing, biomedicine, robot and so on. Therefore, 3D object recognition has broad application prospects. 3D object recognition methods can be roughly divided into two categories: global feature based recognition method and local feature based recognition method. Because of the global feature based recognition method, 3D object recognition method can be divided into two categories: global feature based recognition method and local feature based recognition method. Ignoring some local information about the object, So in some cluttered, occluded scenes, they can't be recognized very well. The 3D object recognition method based on local features can be divided into three stages: detection of 3D feature points. The description of feature points and the matching of local surfaces. In this paper, the method of 3D object recognition based on local features is deeply studied around these three core stages. In this paper, a multi-scale feature extraction algorithm based on scattered point cloud is proposed to solve the problem of object scale invariance and the sensitivity of traditional algorithm to noise. By changing the size of the local neighborhood to construct the scale space for multi-scale analysis, the surface variation value is calculated by the covariance analysis of the local neighborhood at different scales. At the same time, a point signature method based on shape index value is introduced, which enhances the robustness to noise. In this paper, a geometric covariance descriptor is proposed, in which the normal vector angle and distance between feature points and neighborhood points are used to construct the covariance matrix to describe the feature points. This descriptor not only has strong description ability, but also has rotation translation invariance, is not sensitive to noise, and has strong robustness to point cloud sampling density. Because there are some mismatch pairs after nearest neighbor matching of feature points, this paper introduces the canonical correlation analysis into the feature points' mismatch elimination, and finally gets a better matching effect.
【學(xué)位授予單位】：中北大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：TP391.41

【相似文獻(xiàn)】

相關(guān)期刊論文 前10條

1 曾接賢,魯宇明,黎明,儲(chǔ)s，

本文編號(hào)：1512098