【摘要】：隨著工業(yè)4.0的全面發(fā)展,機(jī)器人已經(jīng)不單純是機(jī)械領(lǐng)域的代表,人工智能、機(jī)器視覺和云計(jì)算的發(fā)展,使得機(jī)器人的功能和應(yīng)用場景更加豐富,人類社會(huì)對于智能機(jī)器人的需求也變得多樣化,在這些眾多的需求中,對于機(jī)器人能夠在空間中自主定位并對空間進(jìn)行三維重構(gòu)這一項(xiàng)需求來說,是很多功能的基礎(chǔ),而且直接決定了機(jī)器人能否出現(xiàn)在人類未來的生活中。本文所研究的方向就是基于RGB-D相機(jī)的機(jī)器人能夠在空間中進(jìn)行自主定位的同時(shí),來完成對場景地圖的重建。為了降低噪聲對系統(tǒng)的影響使輸出結(jié)果更加準(zhǔn)確并提高系統(tǒng)的穩(wěn)定性,本文主要從以下幾個(gè)方面進(jìn)行研究:首先對RGB-D所拍攝的圖像進(jìn)行分析,得出噪聲在深度相中的分布形況,然后根據(jù)噪聲的分布,為了降低噪聲對運(yùn)動(dòng)估計(jì)的影響,提出了一種基于平面化參數(shù)的特征點(diǎn)降噪提取方法,該種方法能夠利用平面化參數(shù)將特征點(diǎn)的深度值進(jìn)行修正,從而降低噪聲的影響,最后獲得一個(gè)相對穩(wěn)定的運(yùn)動(dòng)估計(jì),使系統(tǒng)的魯棒性得到提高,同時(shí)給出了與傳統(tǒng)提取法的比較實(shí)驗(yàn)。然后,對SLAM(Simultaneous Localization And Mapping)非參數(shù)化方程進(jìn)行了討論,根據(jù)觀測數(shù)據(jù)的形式導(dǎo)出了其概率模型,針對概率模型中的最大似然估計(jì)導(dǎo)出了最小二乘方程,將原問題轉(zhuǎn)化為非線性問題。同時(shí),在求解非線性方程的過程中,本文討論了其增量方程的稀疏性,稀疏性能夠保證方程的求解效率,滿足了系統(tǒng)的實(shí)時(shí)性要求。其次,本文在系統(tǒng)中加入了回環(huán)檢測模塊,回環(huán)檢測模塊的作用是能夠降低定位算法中產(chǎn)生的累積誤差,這種累積誤差是只考慮了連續(xù)時(shí)間上的數(shù)據(jù)關(guān)聯(lián)而造成的,而回環(huán)檢測將空間中的約束考慮進(jìn)來,通過對同一場景不同時(shí)刻所拍攝的圖像進(jìn)行匹配,提供新的約束來達(dá)到消除累積誤差的目的。最后,根據(jù)最后優(yōu)化得到的運(yùn)動(dòng)軌跡數(shù)據(jù),來對場景進(jìn)行三維重建,本文中展示了兩種建圖方式:點(diǎn)云地圖和八叉樹地圖。根據(jù)使用場景和應(yīng)用需求的不同,對這兩種構(gòu)建地圖的方式進(jìn)行了對分析和討論。
[Abstract]:With the development of industry 4.0, robot is not only the representative of mechanical field, but also the development of artificial intelligence, machine vision and cloud computing. The demands of human society for intelligent robots have also become diversified. Among these many needs, the need for robots to be able to independently locate in space and reconstruct space in three dimensions is the basis of many functions. And directly determines whether the robot can appear in the future of human life. The research direction of this paper is that the robot based on RGB-D camera can locate its own position in space and complete the reconstruction of scene map at the same time. In order to reduce the effect of noise on the system and improve the stability of the system, this paper mainly studies the following aspects: firstly, the image taken by RGB-D is analyzed, and the distribution of noise in the depth phase is obtained. Then, according to the distribution of noise, in order to reduce the influence of noise on motion estimation, a method of feature point denoising based on plane parameters is proposed, which can be used to modify the depth of feature point. Finally, a relatively stable motion estimation is obtained and the robustness of the system is improved. At the same time, a comparison experiment with the traditional extraction method is given. Then, the nonparametric equations of slam (simulated localization And mapping) are discussed, its probability model is derived according to the form of observed data, and the least square equation is derived for the maximum likelihood estimation in the probabilistic model, and the original problem is transformed into a nonlinear problem. At the same time, in the process of solving nonlinear equations, the paper discusses the sparsity of the incremental equations, which can guarantee the efficiency of solving the equations and meet the real-time requirements of the system. Secondly, this paper adds the loop detection module to the system. The function of the loop detection module is to reduce the cumulative error in the localization algorithm, which is caused by only considering the data association in the continuous time. Loop detection takes space constraints into account and provides new constraints to eliminate cumulative errors by matching images taken at different times in the same scene. Finally, according to the motion track data obtained from the final optimization, 3D reconstruction of the scene is carried out. In this paper, two kinds of mapping methods are shown: point cloud map and octree map. According to the different usage scenarios and application requirements, this paper analyzes and discusses the two ways of building maps.
【學(xué)位授予單位】：哈爾濱工業(yè)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;TP242

【參考文獻(xiàn)】

相關(guān)期刊論文 前5條

1 劉浩敏;章國鋒;鮑虎軍;;基于單目視覺的同時(shí)定位與地圖構(gòu)建方法綜述[J];計(jì)算機(jī)輔助設(shè)計(jì)與圖形學(xué)學(xué)報(bào);2016年06期

2 夏天維;侯翔;;基于自適應(yīng)Kalman濾波的機(jī)器人運(yùn)動(dòng)目標(biāo)跟蹤算法[J];計(jì)算機(jī)測量與控制;2015年01期

3 李沫;郝偉博;范哲意;劉志文;;一種改進(jìn)的粒子濾波和Mean Shift聯(lián)合跟蹤算法[J];中國電子科學(xué)研究院學(xué)報(bào);2013年06期

4 梁明杰;閔華清;羅榮華;;基于圖優(yōu)化的同時(shí)定位與地圖創(chuàng)建綜述[J];機(jī)器人;2013年04期

5 程建;周越;蔡念;楊杰;;基于粒子濾波的紅外目標(biāo)跟蹤[J];紅外與毫米波學(xué)報(bào);2006年02期

相關(guān)博士學(xué)位論文 前1條

1 李立春;基于無人機(jī)序列成像的地形重建及其在導(dǎo)航中的應(yīng)用研究[D];國防科學(xué)技術(shù)大學(xué);2009年

相關(guān)碩士學(xué)位論文 前1條

1 王健;基于單目視覺的機(jī)器人焊縫識(shí)別與軌跡規(guī)劃[D];上海交通大學(xué);2012年

，

本文編號(hào)：2128622