循跡搬運(yùn)機(jī)器人--畢業(yè)設(shè)計(jì)（優(yōu)秀含CAD圖紙+設(shè)計(jì)說(shuō)明書(shū)）

循跡搬運(yùn)機(jī)器人--畢業(yè)設(shè)計(jì)（優(yōu)秀含CAD圖紙+設(shè)計(jì)說(shuō)明書(shū)）,搬運(yùn),機(jī)器人,畢業(yè)設(shè)計(jì),優(yōu)秀,優(yōu)良,cad,圖紙,設(shè)計(jì),說(shuō)明書(shū),仿單 畢業(yè)論文（設(shè)計(jì)） 外文翻譯 題 目： 循跡搬運(yùn)機(jī)器人設(shè)計(jì) 系部名稱(chēng)： 機(jī)械工程系 專(zhuān)業(yè)班級(jí)： xxxxx 學(xué)生姓名： xxxxx 學(xué) 號(hào)： xxxxx 指導(dǎo)教師： xxxxx 教師職稱(chēng)： 教 授 2015 年 3 月 13 日 中原工學(xué)院信息商務(wù)學(xué)院外文翻譯 摘要 本章介紹了視覺(jué)系統(tǒng)對(duì)移動(dòng)機(jī)器人跟蹤和控制的一個(gè)完整的主題。它包括用于估計(jì)改善不良的工作條件，例如噪聲，攝像機(jī)鏡頭的失真和非均勻照明的位置及方向的估計(jì)方法和移動(dòng)機(jī)器人全局視覺(jué)系統(tǒng)。 視覺(jué)系統(tǒng)的基本動(dòng)作被分為兩個(gè)步驟。在第一個(gè)中，輸入圖像被掃描的像素分為有限數(shù)量。與此同時(shí)，分割算法是用來(lái)尋找的相應(yīng)的區(qū)域?qū)儆谝活?lèi)。在第二步驟中，所有的區(qū)域進(jìn)行檢查。精選的那些是所觀察到的物體的一部分是通過(guò)簡(jiǎn)單的邏輯程序裝置制成。所使用的方法的新穎性，重點(diǎn)是完成可能的對(duì)象的位置要估計(jì)所需要的處理時(shí)間的優(yōu)化。 進(jìn)一步關(guān)于一種方法，以提高在惡劣的工作條件已經(jīng)存在的視覺(jué)系統(tǒng)性能提出。一些基礎(chǔ)知識(shí)和解決方案，在視覺(jué)系統(tǒng)設(shè)計(jì)的移動(dòng)機(jī)器人跟蹤伴隨的問(wèn)題給出。除了用于過(guò)濾和改進(jìn)識(shí)別噪聲數(shù)據(jù)而劣化的性能的主要因素被處理，即非均勻照明和攝像機(jī)鏡頭畸變的方法。對(duì)于前者的問(wèn)題區(qū)域和它的起源都集中在和通過(guò)施加由照明平原定義乘法組件及其補(bǔ)償?shù)娜芤航o出。后者包括兩個(gè)步驟。首先，徑向鏡頭畸變的基本面進(jìn)行了討論。其核查建議的解決方案是通過(guò)鏡頭投影的幾何模型來(lái)實(shí)現(xiàn)的。第二步驟涵蓋透視失真從照相機(jī)的傾斜始發(fā)。為它的校正被施加消失點(diǎn)檢測(cè)的有效且可靠的方法。如果實(shí)施以適當(dāng)?shù)姆绞剑@兩種校正方法有助于視覺(jué)系統(tǒng)的性能。 所提出的方法應(yīng)用在機(jī)器人足球比賽試驗(yàn)證實(shí)。機(jī)器人足球比賽是一個(gè)快速的動(dòng)態(tài)博弈，因此需要一個(gè)有效的和強(qiáng)大的視覺(jué)系統(tǒng)。為了提高足球機(jī)器人視覺(jué)系統(tǒng)提出的攝像機(jī)標(biāo)定和照度不均勻校正算法的實(shí)現(xiàn)結(jié)果。鏡頭校正方法成功地校正由鏡頭引起的失真，從而實(shí)現(xiàn)更精確的目標(biāo)位置估計(jì)。光照補(bǔ)償提高魯棒性不規(guī)則和不均勻的照明，幾乎總是存在于現(xiàn)實(shí)條件下。  1引言 使用彩色攝像機(jī)的運(yùn)動(dòng)目標(biāo)檢測(cè)的方法很多。然而，根據(jù)顏色信息的視覺(jué)系統(tǒng)被證明是更簡(jiǎn)單，健壯和比大多數(shù)如[3,5,13,16]表示其它識(shí)別方法更快。 Sargent等人 [13]開(kāi)發(fā)了一種快速實(shí)時(shí)視覺(jué)系統(tǒng)與一個(gè)特殊的硬件加速的系統(tǒng)，這才有意義，該系統(tǒng)對(duì)軟件優(yōu)化或加速度有很大的幫助。移動(dòng)對(duì)象的一個(gè)更可靠的視覺(jué)跟蹤可以通過(guò)使用穩(wěn)健統(tǒng)計(jì)和概率分布來(lái)實(shí)現(xiàn)。后者的一個(gè)很好的例子給出了Bradski[2]實(shí)現(xiàn)了基于顏色的人臉跟蹤。 Bruce等[3]通過(guò)高效顏色分割裝置和一個(gè)兩通連通區(qū)域判定算法建議用于移動(dòng)機(jī)器人快速視覺(jué)系統(tǒng)。在機(jī)器人足球視覺(jué)設(shè)計(jì)的另一個(gè)重要貢獻(xiàn)是由惠氏等人介紹。 [16]，以提供給不同的操場(chǎng)光照條件的魯棒性特別考慮。大多數(shù)方法嘗試，以圖像的像素分類(lèi)成的預(yù)定義號(hào)碼之一。最常見(jiàn)的有：線(xiàn)性顏色的閾值，K近鄰分類(lèi)，神經(jīng)網(wǎng)絡(luò)為基礎(chǔ)的分類(lèi)器，分類(lèi)樹(shù)和概率方法[10,1,8]。 本章介紹了當(dāng)前對(duì)象的位置和方向在操場(chǎng)上估計(jì)的全局視覺(jué)系統(tǒng)設(shè)計(jì)。我們感興趣的是在MiroSot類(lèi)足球機(jī)器人上沒(méi)有位置傳感器。因此，一個(gè)準(zhǔn)確的和快速的全球視野，必須設(shè)計(jì)用于機(jī)器人控制和導(dǎo)航中的部分控制的，動(dòng)態(tài)變化的環(huán)境中。當(dāng)設(shè)計(jì)的視覺(jué)系統(tǒng)，以下要求必須完成： ?計(jì)算效率， ?高可靠性， ?良好的精度， ?魯棒性噪聲，非均勻照明和不同的配色方案。 最后一個(gè)特性是必不可少的系統(tǒng)功能以及當(dāng)參與比賽[16]在不同條件下使用它。 在本文中，隨著不斷的閾值和回步算法，快速的方式呈現(xiàn)，其中特別關(guān)注了效率方面。該閾值可被表示為在三維彩色空間框。這些閾值是由離線(xiàn)學(xué)習(xí)來(lái)確定。如果一個(gè)輸入像素的顏色落在一個(gè)預(yù)定義的盒子，那是屬于這個(gè)箱子相關(guān)的類(lèi)。在第一步驟之后是其中屬于一個(gè)類(lèi)（一個(gè)連接區(qū)域）的像素區(qū)別標(biāo)記的第二步驟。以獲得所有完全連接區(qū)域的主要目的，是應(yīng)用一步算法。這兩個(gè)步驟都只有一個(gè)掃描的圖像。然后邏輯部分和一個(gè)簡(jiǎn)單的優(yōu)化方法被用來(lái)從先前生成的那些選擇適當(dāng)?shù)膮^(qū)域?qū)儆谝活?lèi)。這樣的邏輯是操場(chǎng)上的物體的位置和方向估計(jì)。以改善視覺(jué)系統(tǒng)相機(jī)校準(zhǔn)和非均勻照明校正算法被實(shí)現(xiàn)結(jié)果。由相機(jī)透鏡導(dǎo)致的，從而實(shí)現(xiàn)更準(zhǔn)確和精確的目標(biāo)位置估計(jì)，而后者提高了魯棒性不規(guī)則照明和非均勻照明條件。 改善不良的工作條件下與已經(jīng)存在的視覺(jué)系統(tǒng)性能的嘗試接著呈現(xiàn)。兩個(gè)主要因素對(duì)性能產(chǎn)生不利影響，處理：非均勻照明和攝像機(jī)鏡頭失真。對(duì)于前者，重點(diǎn)放在問(wèn)題區(qū)域和它的起源[6]，具有由給定的照明平面中定義的乘法部件的應(yīng)用裝置，用于它的補(bǔ)償?shù)娜芤?。后者包括兩個(gè)步驟。首先，在徑向鏡頭畸變基本面上進(jìn)行了討論[9,14]。對(duì)于其驗(yàn)證建議的解決方案是由透鏡投射[11]一個(gè)幾何模型的手段來(lái)實(shí)現(xiàn)。第二步驟涵蓋透視失真從照相機(jī)的傾斜始發(fā)。對(duì)于其改正，消失點(diǎn)檢測(cè)[4,12]的高效和可靠的方法被應(yīng)用。所提出的方法的適用性在機(jī)器人足球測(cè)試床確認(rèn)。為了提高機(jī)器人足球視覺(jué)系統(tǒng)[5]，這兩個(gè)建議的攝像機(jī)標(biāo)定和非均勻照明校正算法被實(shí)現(xiàn)。 本章安排如下。在第2節(jié)系統(tǒng)的簡(jiǎn)要概覽，用于像素分類(lèi)的方法在部分解釋。第3、第4節(jié)的重點(diǎn)是算法的圖像分割和區(qū)域標(biāo)記。該算法為對(duì)象估計(jì)說(shuō)明在第5節(jié)，第6節(jié)恢復(fù)數(shù)據(jù)過(guò)濾，攝像機(jī)標(biāo)定和非均勻的光量校正的實(shí)現(xiàn)。得到的實(shí)驗(yàn)結(jié)果顯示在第7節(jié)。結(jié)論和一些想法，本章最后的結(jié)論和對(duì)未來(lái)工作的一些想法。 2系統(tǒng)概述 所提出的視覺(jué)系統(tǒng)在機(jī)器人足球比賽設(shè)置展示。足球機(jī)器人的設(shè)置，如圖1所示，由十類(lèi)（MiroSot機(jī)器人形成兩隊(duì)）大小為7.5立方厘米，一個(gè)長(zhǎng)方形的操場(chǎng)面積2.2×1.8米，數(shù)字彩色攝像機(jī)索尼dfw-v500，和個(gè)人計(jì)算機(jī)的奔騰4。程序的視覺(jué)部分處理輸入的圖像，分辨率為640×480像素，確定位置和方向的機(jī)器人和球的位置。每個(gè)機(jī)器人有兩個(gè)方形色塊（圖2）。一個(gè)是球隊(duì)的顏色和其他識(shí)別色標(biāo)。據(jù)FIRA（國(guó)際機(jī)器人足球聯(lián)合會(huì)）的規(guī)則，球隊(duì)的顏色是藍(lán)色或黃色，球必須是橙色和識(shí)別色可以是任何顏色除了團(tuán)隊(duì)和球的顏色。視覺(jué)算法在操場(chǎng)上以它們的顏色和形狀的考慮上找到對(duì)象。如果一個(gè)輸入像素的顏色落在一個(gè)預(yù)定義的盒子（定義的閾值），它是屬于這個(gè)箱子相關(guān)的類(lèi)。閾值是在三維顏色空間的盒子。屬于一個(gè)類(lèi)的像素（連接區(qū)域）進(jìn)行獨(dú)特的標(biāo)記。邏輯部分和一個(gè)簡(jiǎn)單的優(yōu)化方法從先前生成的，選擇合適的地區(qū)。最后，程序的控制部分計(jì)算的線(xiàn)性和角速度，使機(jī)器人應(yīng)該在下一采樣時(shí)刻根據(jù)場(chǎng)上的形勢(shì)。這些參考轉(zhuǎn)速是通過(guò)無(wú)線(xiàn)連接發(fā)送給機(jī)器人，他們開(kāi)始根據(jù)接收到的命令移動(dòng)。上述循環(huán)重復(fù)30次每秒。 無(wú)線(xiàn)電發(fā)射機(jī) 籌略 標(biāo)定攝像機(jī) 計(jì)算機(jī)視覺(jué) 估計(jì)對(duì)象 確定連接區(qū)域 離線(xiàn)學(xué)習(xí)/初始化 陰影校正 多閾值 不斷的閾值 圖1：系統(tǒng)概述 識(shí)別 合作 圖2：機(jī)器人色標(biāo) 8結(jié)論 本章地址是傷腦筋的機(jī)器人足球社區(qū)問(wèn)題；往往是通過(guò)瑣碎的問(wèn)題。然而，有對(duì)問(wèn)題的有效解決，文獻(xiàn)較少，這往往是讓人失望，球隊(duì)希望追求其他的問(wèn)題在足球領(lǐng)域的源（如AI控制）。 建立一個(gè)足球比賽中的移動(dòng)機(jī)器人的目的，快速和強(qiáng)大的視覺(jué)系統(tǒng)的一個(gè)例子。特別考慮到工作和魯棒性問(wèn)題的優(yōu)化計(jì)算。后者是通過(guò)對(duì)圖像質(zhì)量的改善，如非均勻光照和鏡頭畸變校正方法包含放心。魯棒性是通過(guò)具有時(shí)效性的算法使圖像處理進(jìn)一步實(shí)現(xiàn)全球。相反，一些視覺(jué)系統(tǒng)的機(jī)器人足球隊(duì)雇傭當(dāng)?shù)氐膱D像處理獲得的視覺(jué)系統(tǒng)所需的幀速率的應(yīng)用。這些算法的主要缺點(diǎn)是一個(gè)或多個(gè)對(duì)象的損失（機(jī)器人或球）因?yàn)橐恍┎豢深A(yù)知的原因（光照下，碰撞，錯(cuò)誤）。局部搜索區(qū)域必須被增加直到找到對(duì)象，這導(dǎo)致在較大的和不規(guī)則的采樣時(shí)間。這不可能發(fā)生的全球圖像處理。然而，該方法的缺點(diǎn)會(huì)出現(xiàn)如果大量（超過(guò)15）的不同色塊，必須遵循。一些色塊可以成為相機(jī)的圖像可能會(huì)導(dǎo)致錯(cuò)誤的對(duì)象估計(jì)很相似。這個(gè)問(wèn)題將在目標(biāo)跟蹤算法納入未來(lái)的工作處理。 進(jìn)一步的方法建立一個(gè)更強(qiáng)大和精確的移動(dòng)機(jī)器人的不良照明和攝像機(jī)鏡頭畸變條件下跟蹤視覺(jué)系統(tǒng)。為了提高機(jī)器人視覺(jué)跟蹤結(jié)果，提出了攝像機(jī)標(biāo)定和照度不均勻校正算法。前校正由鏡頭引起的失真，從而實(shí)現(xiàn)更準(zhǔn)確和精確的目標(biāo)位置估計(jì)，后者提高了魯棒性不規(guī)則的照明和非均勻光照條件。所建議的解決方案的適用性在機(jī)器人足球比賽中證明，任何不正確或不準(zhǔn)確估計(jì)機(jī)器人和球的位置導(dǎo)致的游戲（除了策略控制算法完善）。視覺(jué)系統(tǒng)的魯棒性是通過(guò)攝像機(jī)標(biāo)定算法提高。建議的程序?yàn)殛幱靶Ｕ蛔C明是有用的，在光照條件下或多或少保持不變?cè)谟螒?。程序也假定固定攝像機(jī)視圖，在中央視覺(jué)系統(tǒng)。在一般的移動(dòng)機(jī)器人，并不總是符合這些條件。如果光照條件的變化，在跟蹤過(guò)程中，一個(gè)適應(yīng)機(jī)制更有效的方法應(yīng)。這將是進(jìn)一步研究解決。優(yōu)化算法，使視覺(jué)系統(tǒng)可以用于實(shí)時(shí)應(yīng)用在不規(guī)則的照明和攝像機(jī)畸變的魯棒性是很重要的。 本文摘自《VISION SYSTEM DESIGN FOR MOBILE ROBOT TRACKING》。  Abstract This chapter introduces a complete thematic of vision system for mobile robot tracking and control. It consists of a global vision system for estimation of positions and orientations of mobile robots and methods for improvement of bad operating conditions such as noise, camera lens distortion and non-uniform illumination. The basic operation of a vision system is divided into two steps. In the first, the incoming image is scanned and pixels are classified into a finite number of classes. At the same time, a segmentation algorithm is used to find the corresponding regions belonging to one of the classes. In the second step, all the regions are examined. A selection of the ones that are a part of the observed object is made by means of simple logic procedures. The novelty of the used approach is focused on optimization of the processing time needed to finish the estimation of possible object positions. Further on an approach to improve an already existing vision system performance under bad operating conditions is presented. Some fundamentals and solutions to accompanying problems in vision system design for mobile robot tracking are presented. Besides methods for filtering and improvement of identified noisy data the two main factors which deteriorate the performance are dealt with, namely, non-uniform illumination and camera lens distortion. For the former the problem area and its origins are focused on and a solution for its compensation by applying multiplicative component defined by illumination plain is given. The latter consists of two steps. In the first, radial lens distortion fundamentals are discussed. The suggested solution for its verification is realized by a geometry model of lens projection. The second step covers the perspective distortion originating from the tilt of the camera. For its correction an efficient and robust method of vanishing point detection is applied. Both correction methods contribute to a vision system performance if implemented in the appropriate manner. Applicability of the presented approaches is confirmed on a robot soccer test bed. Robot soccer is a fast dynamic game and therefore needs an efficient and robust vision system. To improve the results of the robot soccer vision system the proposed camera calibration and non-uniform illumination correction algorithm are implemented. The lens correction method successfully corrects the distortion caused by the camera lens, thus achieving a more accurate and precise estimation of the object position. The illumination compensation improves robustness to irregular and non-uniform illumination which is nearly always present in real conditions. 1 Introduction There are many ways of detecting moving objects using color cameras. However, the vision systems based on color information proved to be more simple, robust and faster than most of other recognition methods as stated in [3,5,13,16]. Sargent et al. [13] developed a fast real-time vision system with the aid of a special hardware accelerated system, which only makes sense when software optimizations or accelerations are not possible. A more reliable vision tracking of moving objects can be achieved by using robust statistics and probability distributions. A good example of the latter is given in the color-based face tracking implemented by Bradski [2]. Bruce et al. [3] suggested a fast vision system for mobile robots by means of efficient color segmentation and a two-pass connected region determination algorithm. Another important contribution to the robot soccer vision design was introduced by Wyeth et al. [16], with special consideration given to the robustness of varying playground illumination conditions. Most of the approaches try to classify the pixels of an image into one of a predefined number of classes. The most common are: linear color thresholding, K-nearest neighbor classification, neural net-based classifiers, classification trees and probabilistic methods [10,1,8]. The chapter presents a design of a global vision system for estimating current object positions and orientations on the playground. The MiroSot category soccer robots we are interested in are without on-board position sensors. Thus a precise and fast global vision has to be designed for robots control and navigation in a partially controlled, dynamically changing environment. When designing the vision system, the following requirements have to be accomplished: ? computational efficiency, ? high reliability, ? good precision, and ? robustness to noise, non-uniform illumination and different color schemes. The last characteristic is essential for the system to function well when using it under different conditions present at competitions [16]. In this paper, a fast approach with constant thresholding and back-stepping algorithm is presented, where a special attention is given to the efficiency aspect. The thresholds can be presented as boxes in 3-dimensional color spaces. These thresholds are determined by means of off-line learning. If an incoming pixel color falls inside one of the predefined boxes, then it is classified as belonging to the class associated with this box. This first step is followed by the second step where the pixels belonging to one class (a connected region) are distinctively labeled. With the main purpose of obtaining all fully connected regions, a back-stepping algorithm is applied. Both steps are done with just one scan of the image. Then the logic part and a simple optimization method are employed to select the proper regions from the previously generated ones. After this logic the positions and orientations of the objects on the playground are estimated. To improve results of the vision system the camera calibration and non-uniform illumination correction algorithm are implemented. The former corrects distortion caused by the camera lens, thus achieving a more accurate and precise objects positions estimation, while the latter improves robustness to irregular illumination and non-uniform illumination conditions. An attempt to improve the already-existing vision systems performance under poor operating conditions is next presented. Two main factors, which adversely affect performance, are dealt with: non-uniform illumination and camera lens distortion. For the former, the focus is placed on the problem area and its origins [6], with a solution for its compensation by means of the application of a multiplicative component defined by an illumination plane given. The latter consists of two steps. In the first, radial lens distortion fundamentals are discussed [9,14]. The suggested solution for its verification is realized by means of a geometric model of lens projection [11]. The second step covers the perspective distortion originating from the tilt of the camera. For its correction, an efficient and robust method of vanishing point detection [4,12] is applied. The applicability of the presented approaches is confirmed on the robot soccer test bed. To improve the results of the robot soccer vision system [5], both the proposed camera calibration and non-uniform illumination correction algorithm are implemented. The chapter is organized as follows. In section 2 a brief overview of the system is given. The method used for pixel classification is explained in section 3. Section 4 focuses on the algorithms for image segmentation and region labeling. The algorithm for object estimation is illustrated in section 5. Section 6 resumes the data filtering, camera calibration and non-uniform illumination correction implementation. Obtained experimental results are shown in section 7. The chapter ends with conclusions and some ideas for future work. 2 System Overview The presented vision system is demonstrated on a robot soccer set-up. The soccer robot set-up, Fig. 1, consists of ten MiroSot category robots (forming two teams) of size 7.5 cm cubed, a rectangular playground of size 2.2×1.8 m, a digital color camera Sony DFW-V500, and a personal computer Pentium 4. The vision part of the program processes the incoming images, of a resolution of 640×480 pixels, to identify the positions and orientations of the robots and the position of the ball. Each robot has two square-shaped color patches (Fig. 2). One is the team color and the other is the identification color patch. According to FIRA (Federation of International Robot-soccer Association) rules, the team color must be blue or yellow, the ball must be orange and identification colors can be any color except the team and ball color. The vision algorithm finds objects on the playground by taking their color and shape into consideration. If an incoming pixel color falls inside one of the predefined boxes (defined by thresholds), it is classified as belonging to the class associated with this box. The thresholds are presented as boxes in three-dimensional color spaces. The pixels belonging to one class (a connected region) are then distinctively labeled. The logic part and a simple optimization method are employed to select the proper regions from the previously generated ones. Finally, the control part of the program calculates the linear and angular speeds, that the robots should have in the next sample time according to the current situation on the playground. These reference speeds are sent to the robots by a wireless connection and they start moving according to the received commands. The above-mentioned cycle repeats itself 30 times per second. 8 Conclusion The issues the chapter address are vexing ones for the robot soccer community; issues that are often passed as trivial. However, there is little literature on effective solutions to the problems, which is often source of frustration to teams who wish to purse other issues in the soccer domain (such as AI and control). An example of establishing a fast and robust vision system for the purpose of mobile robots in soccer game is presented. Special consideration is given to optimization of computational work and robustness issues. The latter are assured by inclusion of methods for image quality improvement such as correction of nonuniform illumination and camera lens distortions. Robustness is further achieved by time-efficient algorithms which enable global image processing. Contrary, some vision systems used by other robot soccer teams employ local image processing to obtain the desired frame rate of the vision system. The major disadvantage of these algorithms is loss of one or more objects (robots or ball) because of some unpredicted reasons (lightening conditions, collisions, bugs). The local search areas have to be increased until objects are found, which results in larger and irregular sample time. This could not happen with global image processing. However, disadvantage of the presented approach can appear if a large number (more than 15) of different color patches have to be followed. Some of color patches could then become quite similar on camera image which could result in wrong objects estimation. The problem will be dealt with in the future work by inclusion of object tracking algorithms. Further on an approach towards establishing a more robust and accurate vision system for mobile robot tracking under poor illumination and camera lens distortion conditions is presented. To improve the results of visual robot tracking, a camera calibration and non-uniform illumination correction algorithm are suggested. The former corrects the distortion caused by the camera lens, thus achieving a more accurate and precise estimation of object position, while the latter improves robustness to irregular illumination and non-uniform illumination conditions. The applicability of the suggested solutions is demonstrated in a robot soccer game, where any incorrect or inaccurately estimated robot or ball position results in poor game-play (apart from perfection of the strategy control algorithm). The robustness of the vision system is therefore improved by means of camera calibration algorithms. The suggested procedure for shading correction proved useful when the illumination conditions remained more or less unchanged during the game. The procedure presented also assumes fixed camera view, as in central vision systems. In general mobile robotics, these conditions are not always met. If illumination conditions change during tracking, a more robust approach with an adaptation mechanism should be applied. This will be addressed in further research. The optimized algorithms presented enable the vision system to be used in real-time applications where robustness to irregular illumination and camera distortions are important. 10 教師評(píng)語(yǔ) 教師簽名： 2015 年 月 日