發(fā)布時(shí)間：2018-05-19 06:29

  本文選題：深度學(xué)習(xí) + 幾何模型��； 參考：《廣東工業(yè)大學(xué)》2017年碩士論文

【摘要】：機(jī)器人利用機(jī)器視覺(jué)進(jìn)行物體抓取是機(jī)器人應(yīng)用領(lǐng)域的熱門(mén)研究之一。目的在于靠機(jī)器視覺(jué)檢測(cè)出被抓取物體的可靠抓取位置和方向,進(jìn)而通過(guò)運(yùn)動(dòng)規(guī)劃算法控制機(jī)械臂完成抓取動(dòng)作。但是在這一領(lǐng)域暫未有普適的方法,特別是在需要同時(shí)考慮可靠性和安全性的成堆物體抓取應(yīng)用。此論文提出一種利用深度學(xué)習(xí)網(wǎng)絡(luò)結(jié)合軟性機(jī)械手幾何模型的方法,用于分析處理安裝在機(jī)器人中部的單個(gè)深度攝像頭所獲取的3維點(diǎn)云數(shù)據(jù),實(shí)現(xiàn)在未知的成堆物體中檢測(cè)出可供機(jī)械臂進(jìn)行可靠安全抓取的物體位置和方向。該論文所提出的方法在考慮成堆物體抓取碰撞檢測(cè)問(wèn)題的同時(shí),并沒(méi)有進(jìn)行像其他論文所常用的圖像分割和物體識(shí)別的處理技術(shù)。特別地,此論文還分析了深度卷積神經(jīng)網(wǎng)絡(luò)的時(shí)間和空間復(fù)雜度,并得出兩個(gè)結(jié)論用于為深度卷積神經(jīng)網(wǎng)絡(luò)的設(shè)計(jì)提供參考。同時(shí),進(jìn)行多個(gè)實(shí)驗(yàn)來(lái)得出不同層數(shù)的神經(jīng)網(wǎng)絡(luò)對(duì)應(yīng)的訓(xùn)練時(shí)間和精度,不同層數(shù)神經(jīng)網(wǎng)絡(luò)進(jìn)行假設(shè)可抓取超平面再篩選的結(jié)果對(duì)比,從而驗(yàn)證前面由理論推出來(lái)的兩個(gè)結(jié)論。經(jīng)上述理論及實(shí)驗(yàn)兩方面分析,為深度學(xué)習(xí)模型的設(shè)計(jì)起到一定的指導(dǎo)性作用。值得一提的是,本論文中深度學(xué)習(xí)模型訓(xùn)練數(shù)據(jù)(所有的假設(shè)超平面)的標(biāo)簽根據(jù)一些標(biāo)準(zhǔn)進(jìn)行自動(dòng)標(biāo)定。也就是說(shuō)這些用于判斷該假設(shè)超平面是否符合抓取條件的標(biāo)準(zhǔn)被深度學(xué)習(xí)網(wǎng)絡(luò)模型所取代。首先,由于軟性機(jī)械手相較于硬性機(jī)械手更具靈活性,為軟性機(jī)械手設(shè)計(jì)合適的幾何模型成為避免碰撞的重點(diǎn),用于在3維點(diǎn)云數(shù)據(jù)空間中搜索符合可抓取條件的抓取點(diǎn)和抓取方向。該幾何模型內(nèi)部必須包含足夠的點(diǎn)云數(shù)據(jù),并且外表面不會(huì)與其他點(diǎn)云數(shù)據(jù)有重合的部分,從而分別保證其抓取的可靠性和安全性。利用幾何模型搜索出來(lái)符合條件的抓取點(diǎn)和抓取方向,合稱之為假設(shè)可抓取超平面。第二,進(jìn)一步考慮抓取可靠性,利用深度學(xué)習(xí)網(wǎng)絡(luò)Mod-Le Net對(duì)搜索出的假設(shè)超平面進(jìn)行分類和排序,以便找出較為可靠的抓取位置和方向。經(jīng)過(guò)Mod-Le Net與支持向量機(jī)技術(shù)的對(duì)比實(shí)驗(yàn),通過(guò)Mod-Le Net的篩選過(guò)后的可抓取超平面的質(zhì)量和可靠性要比通過(guò)支持向量機(jī)的高,而且數(shù)量也相對(duì)較少,也就是說(shuō)在運(yùn)動(dòng)控制方面會(huì)比較節(jié)省時(shí)間。
[Abstract]:Robot object capture using machine vision is one of the hot research in robot application field. The aim of this paper is to detect the position and direction of the captured object reliably by machine vision, and then to control the robot arm to complete the grab by motion planning algorithm. However, there is no universal method in this field, especially in the application of stacks of objects which need to consider both reliability and safety. In this paper, a method of using depth learning network combined with geometric model of soft manipulator is proposed to analyze and process 3D point cloud data obtained by a single depth camera installed in the middle of the robot. The position and direction of the object which can be reliably and safely grasped by the manipulator can be detected in the unknown stacks of objects. The method proposed in this paper does not deal with image segmentation and object recognition as commonly used in other papers, while considering the problem of collision detection. In particular, the time and space complexity of the deep convolution neural network is analyzed, and two conclusions are drawn to provide a reference for the design of the deep convolution neural network. At the same time, several experiments were carried out to obtain the training time and accuracy of neural networks with different layers. The neural networks with different layers were supposed to be able to grasp hyperplane and then compared the results of screening, thus verifying the two conclusions deduced from the theory. Through the theoretical and experimental analysis above, it plays a guiding role in the design of the deep learning model. It is worth mentioning that the labels of the depth learning model training data (all hypothesized hyperplanes) in this paper are automatically calibrated according to some criteria. In other words, these criteria used to determine whether the hyperplane meets the grasping condition are replaced by the depth learning network model. First of all, because soft manipulators are more flexible than rigid manipulators, designing suitable geometric models for soft manipulators becomes the focus of collision avoidance. It is used in 3D point cloud data space to search for grab points and grab directions that meet the grabability criteria. The geometric model must contain enough point cloud data, and the outer surface will not overlap with other point cloud data, so as to ensure the reliability and security of its capture. The geometry model is used to search the grasping points and directions which meet the conditions, which is called hypothetically grabbing hyperplane. Secondly, the grabbing reliability is further considered, and the hypothetical hyperplane is classified and sorted by using the deep learning network (Mod-Le Net) in order to find out the more reliable grab position and direction. Through the contrast experiment of Mod-Le Net and support vector machine, the quality and reliability of grabable hyperplane after Mod-Le Net screening is higher than that of support vector machine, and the quantity is relatively small. In other words, it saves more time in motion control.
【學(xué)位授予單位】：廣東工業(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4條

1 夏群峰;彭勇剛;;基于視覺(jué)的機(jī)器人抓取系統(tǒng)應(yīng)用研究綜述[J];機(jī)電工程;2014年06期

2 韓崢;劉華平;黃文炳;孫富春;高蒙;;基于Kinect的機(jī)械臂目標(biāo)抓取[J];智能系統(tǒng)學(xué)報(bào);2013年02期

3 章軍;須文波;宋浩;;氣動(dòng)柔性關(guān)節(jié)的抓取機(jī)械手的受力分析[J];組合機(jī)床與自動(dòng)化加工技術(shù);2006年08期

4 李志華,鐘毅芳;虛擬手模型及其抓取技術(shù)[J];小型微型計(jì)算機(jī)系統(tǒng);2003年06期

，

本文編號(hào)：1909063