汽車導(dǎo)航系統(tǒng)畢業(yè)論文中英文資料外文翻譯--使用gis數(shù)據(jù)庫(kù)和激光掃描技術(shù)為汽車導(dǎo)航系統(tǒng)獲取路標(biāo)

1、<p><b>　　中文3656字</b></p><p>　　出處：International archives of photogrammetry remote sensing and spatial information sciences, 2003, 34(3/W8): 131-138</p><p>　　使用GIS數(shù)據(jù)庫(kù)和激光掃描技術(shù)為汽車導(dǎo)航系

2、統(tǒng)獲取路標(biāo)</p><p>　　C Brenner, B Elias</p><p><b>　　摘要</b></p><p>　　現(xiàn)在的汽車導(dǎo)航系統(tǒng)以地圖，圖形，以及聲音的形式提供給用戶行駛中的信息，然而他們還遠(yuǎn)遠(yuǎn)不能支持基于道路標(biāo)記的導(dǎo)航，而這也是對(duì)我們來(lái)說(shuō)更簡(jiǎn)單的導(dǎo)航理念，并且這也在不久要實(shí)現(xiàn)的個(gè)人導(dǎo)航系統(tǒng)中占據(jù)重要的位置。為了提供這樣的

3、一種導(dǎo)航，第一步就要識(shí)別恰當(dāng)?shù)牡缆窐?biāo)記——乍一看似乎很簡(jiǎn)單，但是如果考慮到要把覆蓋了歐洲、北美、日本大部分地區(qū)的信息傳輸給數(shù)據(jù)庫(kù)的挑戰(zhàn)，我們就有理由自命不凡了。在這里，我們將講解從已存在的GIS數(shù)據(jù)庫(kù)中獲取道路標(biāo)記的方法。因?yàn)檫@些數(shù)據(jù)庫(kù)大多數(shù)沒(méi)有包含建筑物的高度和視圖信息，我們將展示這些信息怎樣從激光掃描數(shù)據(jù)中分離出來(lái)。</p><p><b>　　1簡(jiǎn)介</b></p>&l

4、t;p>　　1995年在上層階級(jí)的汽車?yán)锲噷?dǎo)航系統(tǒng)就已經(jīng)出現(xiàn)了，而且現(xiàn)在幾乎可以在任何樣式的汽車中找到導(dǎo)航系統(tǒng)。他們是相對(duì)復(fù)雜和成熟的系統(tǒng)可以以數(shù)字地圖，行駛方向圖形，以及行駛中的聲音信息提供路線導(dǎo)航?；厮?980年汽車導(dǎo)航系統(tǒng)開(kāi)始興起的時(shí)候，一些大的問(wèn)題都得到了解決：例如絕對(duì)位置，適合導(dǎo)航的大量地圖的提供，快速算路以及可靠的路線導(dǎo)航。</p><p>　　然而，傳送這些信息的原始概念并沒(méi)有得到較大的改善

5、。聲音的導(dǎo)航仍然用相對(duì)小的提示：（例如 現(xiàn)在向右轉(zhuǎn)），這只涉及到了道路分布的屬性。這不是最理想的，因?yàn)?）路線分布的特征在較大距離的時(shí)候是不可見(jiàn)的，這是因?yàn)樗緳C(jī)受局限的位置以及視角，2）人們最習(xí)慣的導(dǎo)航方式是通過(guò)道路標(biāo)記，也就是沿路中一系列的可識(shí)別可記憶的的圖像的提供。很明顯，作為道路標(biāo)記的建筑物</p><p>　　的提示與聲音提示結(jié)合起來(lái)，將是導(dǎo)航發(fā)展中更人性化的一個(gè)方向，就像我們下邊討論的那樣，這將很好的集

6、成到今天的汽車導(dǎo)航系統(tǒng)中去因?yàn)椴灰馕吨鴮?duì)系統(tǒng)和數(shù)據(jù)結(jié)構(gòu)的大的改動(dòng)。所以，主要的問(wèn)題在于識(shí)別合適的道路標(biāo)記以及估計(jì)他們對(duì)于導(dǎo)航提示的可用性。這里，我們將解釋已存的數(shù)據(jù)庫(kù)怎樣開(kāi)發(fā)以解決第一個(gè)問(wèn)題，而激光瀏覽數(shù)據(jù)庫(kù)將解決后一個(gè)。</p><p>　　2使用激光掃描數(shù)據(jù)集的可見(jiàn)性分析</p><p><b>　　2.1可見(jiàn)性分析</b></p><p>

7、;　　如果我們以來(lái)自激光掃描的DSM上直觀的可見(jiàn)性作為分析的基礎(chǔ)，我們會(huì)做的更好。我們將不會(huì)獲得像當(dāng)初估計(jì)的那樣使建筑物從任何的地點(diǎn)都被清晰地看見(jiàn)。 我們依照下列各項(xiàng)的方式，對(duì)于任何的觀察點(diǎn)的位置和觀看方向定義給予的水平線和垂直的一個(gè)虛擬的照相機(jī)的外部方位看角。 這個(gè)虛擬的照相機(jī)表示為駕駛者的視野。高度起源于DSM本身,然而看角從 GDF 數(shù)據(jù)組中對(duì)應(yīng)的街道的固定方位被獲得。</p><p>　　虛擬圖像的平面然

8、后被光柵過(guò)濾, 每個(gè)圖素定義物體空間的一道光線。所有的光線在物體空間中被追蹤并且用 DSM 決定交集。對(duì)于每次擊中，對(duì)應(yīng)的物體數(shù)據(jù)被試映圖的虛擬圖像查詢獲得。雖然這個(gè)方法與 "光線追蹤" 類似并用在計(jì)算機(jī)圖形方面和平時(shí)假設(shè)計(jì)算中，但是自從我們只在光線的第一個(gè)擊中方面感興趣之后，它實(shí)際上相當(dāng)速，而且DSM只是2.5 D,虛擬圖像的飛機(jī)以此下去可能從底部到頂端被有效率地計(jì)算,向逐漸增加的物體空間進(jìn)軍。</p>

9、<p><b>　　2.2 追蹤可見(jiàn)性</b></p><p>　　在最后的一個(gè)區(qū)段中，單一視野被計(jì)算。 然而,道路標(biāo)記被一個(gè)路線排定指令選擇，而且一定在整個(gè)調(diào)遣期間是看得見(jiàn)的。這可能沿著對(duì)應(yīng)的調(diào)遣定義的軌道追蹤物體的可見(jiàn)性。對(duì)于我們的第一次實(shí)驗(yàn)，我們使用只有一個(gè)自然的附近區(qū)域作為可見(jiàn), 即被虛擬圖像上的飛機(jī)的對(duì)應(yīng)物體覆蓋的區(qū)域。 </p><p>　　

10、圖1顯示了一個(gè)例子。 我們假設(shè)白色的多角形是我們駕駛者使用的軌道。然后問(wèn)題是如果以其他的方法識(shí)別它是一個(gè)道路標(biāo)記的城鎮(zhèn)大廳,是一個(gè)可以被道路標(biāo)記唯一表示的適當(dāng)物體。對(duì)這一次的結(jié)論，我們的運(yùn)算法則是追蹤整個(gè)的軌道,以等距離隔開(kāi)的位置和在軌道旁邊的固定方位產(chǎn)生罰款者眼中虛擬的視野。對(duì)于每個(gè)如此的視野，虛擬圖像中的飛機(jī)上的每個(gè)物體覆蓋的區(qū)域被決定。</p><p>　　圖 2顯示沿著圖1的軌道所有的那些區(qū)域的一種情況。

11、當(dāng)物體出現(xiàn)的時(shí)候, 能產(chǎn)生典型的有遮掩的曲線, 變化比較大和最后消失的就如被途徑人所觀察。 在這種特別的情形中，當(dāng)位置是城鎮(zhèn)門(mén)廳之前并留下狹窄的街道和進(jìn)入廣場(chǎng)的時(shí)候，或視野弄寬的時(shí)候，許多物體在附近看變成規(guī)格為65號(hào)。</p><p>　　為了確定城鎮(zhèn)大廳是否為一個(gè)適當(dāng)?shù)奈矬w, 從圖2上的對(duì)應(yīng)的曲線看，從規(guī)格65到115號(hào)是最大的,也就是城鎮(zhèn)大廳是駕駛者視野中最大的物體。而且，曲線比從規(guī)格13號(hào)開(kāi)始的更大,這意味

12、城鎮(zhèn)大廳是一少部分––大約在進(jìn)入廣場(chǎng)的之前100 公尺處被看見(jiàn)( 可能是決定點(diǎn)) 因此，在這種情況下我們既能查出顯示比較大的物體，也能在駕駛者的視野中將他最早顯示出來(lái)。</p><p>　　圖1 俯視圖上的軌跡線</p><p>　　圖2：基于框架數(shù)字的可見(jiàn)性劃分</p><p><b>　　3數(shù)據(jù)地圖</b></p><p

13、>　　汽車導(dǎo)航系統(tǒng)使用的地圖不僅包含幾何學(xué)和道路網(wǎng)絡(luò)的連接性而且包含了大量的關(guān)于物體，屬性和關(guān)系的附加信息。一個(gè)好的觀點(diǎn)能夠從歐洲的標(biāo)準(zhǔn)ＧＤＦ獲得，舉例來(lái)說(shuō)，（１９９５年３月的地理數(shù)據(jù)文件），其中包括了博物館，戲院，文化中心和市政廳等的信息。</p><p>　　地圖數(shù)據(jù)是被諸如電子地圖的地圖數(shù)據(jù)庫(kù)廠商獲得并通過(guò)交換的方式提供給汽車導(dǎo)航系統(tǒng)生產(chǎn)商的（例如ＧＤＦ）。在那里，它被轉(zhuǎn)換到最后在地圖激光唱碟或數(shù)字化

14、視頻光上被發(fā)現(xiàn)的專有格式。數(shù)據(jù)必須從一種描述形式轉(zhuǎn)換成被汽車導(dǎo)航系統(tǒng)支援的另一種被特殊化的形式，這轉(zhuǎn)變是高度非凡的。時(shí)常，結(jié)構(gòu)和價(jià)值被這個(gè)轉(zhuǎn)換過(guò)程預(yù)先計(jì)算了，目的是為了要減輕航行系統(tǒng)的在線資源 , 例如帶寬和ＣＰＵ時(shí)間。</p><p>　　這個(gè)模塊的其中一部分也是為每個(gè)十字路口產(chǎn)生一個(gè)點(diǎn)陣式，目的是描述所有的可能轉(zhuǎn)向的組合。在汽車導(dǎo)航系統(tǒng)中使用了眾所周知的箭頭符號(hào)來(lái)標(biāo)識(shí)，這就需要所有道路的十字路口的交匯情況將被

15、存儲(chǔ)。</p><p>　　在轉(zhuǎn)向過(guò)程中，對(duì)于帶有路標(biāo)的汽車導(dǎo)航系統(tǒng)的附加信息會(huì)被完整化。在本文中，概括說(shuō)明了是怎樣通過(guò)ＧＤＦ與地圖數(shù)據(jù)和激光掃描數(shù)據(jù)結(jié)合來(lái)確定道路幾何圖形的適合的路標(biāo)，重要的一點(diǎn)是那些附加的數(shù)據(jù)信息僅僅在這個(gè)轉(zhuǎn)換過(guò)程中被使用。在那之后，僅僅是基于路標(biāo)的行使指示還存在，這些是行使指示可能在一種非常緊湊的形式下被編碼，并且要與每一個(gè)十字路口各自的已被存儲(chǔ)在專有地圖格式的數(shù)據(jù)信息相協(xié)調(diào)。因此，路標(biāo)技術(shù)

16、的整合沒(méi)有在現(xiàn)在的汽車導(dǎo)航系統(tǒng)中造成障礙，這些主要問(wèn)題是來(lái)自那些用自動(dòng)或半自動(dòng)方法的指令中的。</p><p>　　4 激光掃描和城市模型</p><p>　　在二十世紀(jì)九十年代，靠空氣傳播的激光掃描作為獲得表面的模型的新方法變得可用。隨后，掃描系統(tǒng)提高了并且指引全球范圍也因?yàn)樽銐虻木茸兊每尚?。今天，靠空氣傳播的激光掃描是一?xiàng)成熟的技術(shù)為大多數(shù)公司提供系統(tǒng)和服務(wù)。掃描很大的區(qū)域是可能的，

17、例如整個(gè)荷蘭已經(jīng)被掃描過(guò)了，德國(guó)的Baden-W¨urttemberg州也正在進(jìn)行掃描，他們中每一個(gè)的面積都超過(guò)了30平方千米。天線激光掃描機(jī)直接地生產(chǎn)地球的表面密集的點(diǎn)云 (Baltsavias et al。,1999). 他們對(duì)獲得密集的都市區(qū)域的數(shù)傳表面模型 (DSMs) 是特別地適當(dāng)?shù)? 如同他們保存跳躍邊緣一樣相當(dāng)好。 大多數(shù)的系統(tǒng)能夠測(cè)量不只有高度, 也有反射系</p><p>　　數(shù), 和

18、首先,最后的或多樣的回行脈沖,他們?cè)试S分開(kāi)樹(shù)形天篷和地面。 (Kraus 和 Rieger,1999)</p><p>　　主要的問(wèn)題是怎樣從激光掃描數(shù)據(jù)組中獲取關(guān)于人造結(jié)構(gòu)的符號(hào)信息，可能和天空的或陸地的圖像聯(lián)合。尤其, 自動(dòng)機(jī)械世代的城市模型是而且仍然是一個(gè)強(qiáng)烈的研究領(lǐng)域, 這個(gè)討論是超過(guò)本文的范圍的。 在這一問(wèn)題上，讀者可以咨詢“Ascona 工作室”的優(yōu)秀的成果。 (Gr¨un et al.,

19、1995, Gr¨un et al., 1997,Baltsavias et al., 2001).</p><p>　　然而，實(shí)質(zhì)性研究努力還是很必要的直到高度自動(dòng)化的物體獲取系統(tǒng)可以可靠地工作。另一方面，三維空間存在的物體信息在今天存在的GIS數(shù)據(jù)庫(kù)中還遠(yuǎn)遠(yuǎn)不是普遍的。所以，在本文中我們將考慮把GIS數(shù)據(jù)庫(kù)和激光掃描DSMs聯(lián)合起來(lái)在一個(gè)圖標(biāo)層上，不明確地重建物體的三維空間的形狀而當(dāng)做分開(kāi)實(shí)體。圖3

20、展示了一個(gè)數(shù)據(jù)資源被用過(guò)的例子，來(lái)自正在激光掃描的DSM，使有規(guī)則到1米的格子，街道的幾何形狀用從一個(gè)GDF數(shù)據(jù)組合的中心線表示，而建筑物的輪廓用從地籍圖上獲得的中心線表示。</p><p><b>　　圖3 激光掃描</b></p><p><b>　　5 結(jié)論及前景</b></p><p>　　在本文中，已經(jīng)概略說(shuō)明路

21、標(biāo)是如何被取得的并且評(píng)估使用已存在的 GIS 和激光掃描數(shù)據(jù)。 至于路標(biāo)的取得，我們已經(jīng)調(diào)查基于顯示突出建筑物的二種不同的方法。 為了評(píng)估導(dǎo)航引導(dǎo)的有用性,我們用了基于來(lái)自激光掃描的 DSM 數(shù)據(jù)的一項(xiàng)可見(jiàn)性分析。 </p><p>　　數(shù)據(jù)挖掘程序必須用真正的數(shù)據(jù)組來(lái)測(cè)試。 如果他們?cè)诂F(xiàn)實(shí)世界中使用適當(dāng)?shù)穆窐?biāo)引導(dǎo)，這個(gè)結(jié)論將會(huì)被證實(shí)。 除此之外，分析程序必須被擴(kuò)展到不同的事物類型 (交通建筑,公園,體育運(yùn)動(dòng)設(shè)備

22、等.) 從 ATKIS 數(shù)據(jù)提取舉例來(lái)說(shuō)明。不同種類事物的數(shù)據(jù)預(yù)處理方法和當(dāng)不同數(shù)據(jù)挖掘運(yùn)算法則被提供到相同數(shù)據(jù)時(shí)產(chǎn)生的問(wèn)題必須被調(diào)查。萃取的路標(biāo)的可靠性不得不通過(guò)質(zhì)量測(cè)試來(lái)決定，目的是為了避免不明確的目標(biāo)誤導(dǎo)用戶。</p><p>　　更多的依靠路線來(lái)決定路標(biāo)的問(wèn)題必須被調(diào)查: 用戶行駛方向和路標(biāo)質(zhì)量可見(jiàn)性的影響。當(dāng)我們只用了 "虛擬的圖像大小" 來(lái)估價(jià)一個(gè)事物的可見(jiàn)性時(shí),有很大的空間來(lái)進(jìn)步

23、。舉例來(lái)說(shuō)，如果一個(gè)事物被它前面或附近的事物擋住了，或者是整個(gè)輪廓的一部分，從虛擬的圖像，就能獲得遠(yuǎn)距離的信息。首先激光掃描測(cè)量的脈搏能夠被整合，目的是為了獲得一個(gè)比較好的由樹(shù)導(dǎo)致阻塞的近似值。DSM 也可能被用來(lái)提供萃取的附加信息，例如，小塔被它前面的大建筑物擋住這個(gè)信息將被確定。跟蹤可見(jiàn)性的執(zhí)行使用等距離的時(shí)間取樣來(lái)代替空間取樣，這是基于車輛在臨近交叉路口的速度的。最后，存在于GDF數(shù)據(jù)中的POI數(shù)據(jù)被使用到可見(jiàn)性分析的擴(kuò)展是非常有

24、趣的。</p><p><b>　　附件2：外文原文</b></p><p>　　EXTRACTING LANDMARKS FOR CAR NAVIGATION SYSTEMS USING EXISTING GIS DATABASES AND LASER SCANNING</p><p><b>　　ABSTRACT</b>

25、;</p><p>　　Today’s car navigation systems provide driving instructions in the form of maps, pictograms, and spoken language. However, they are so far not able to support landmark-based navigation, which is t

26、he most natural navigation concept for humans and which also plays an important role for upcoming personal navigation systems. In order to provide such a navigation, the first step is to identify appropriate landmarks –

27、a task that seems to be rather easy at first sight but turns out to be quite pretent</p><p>　　1 INTRODUCTION</p><p>　　Modern car navigation systems have been introduced in 1995 in upper class ca

28、rs and are now available for practically any model. They are relatively complex and mature systems able to provide route guidance in form of digital maps, driving direction pictograms,and spoken language driving instruct

29、ions (Zhao, 1997).Looking back to the first beginnings in the early 1980s, many nontrivial problems have been solved such as absolute positioning, provision of huge navigable maps, fast routing and reliab</p><

30、p>　　However, the original concept of delivering the instructions has not changed very much. Still, spoken language instructions use a relatively small set of commands (like ’turn right now’), which only refer to prope

31、rties of the street network. This is not optimal, since i) features of the street network typically are not visible from a greater distance due to the low driver position and small observing angle, and ii) the most natur

32、al form of navigation for humans is the navigation by landmarks, i.</p><p>　　2VISIBILITY ANALYSIS USING LASER SCANNING DATASETS</p><p>　　2.1 Visibility Analysis</p><p>　　we can do b

33、etter if we base the visibility analysis directly on the DSM from laser scanning. We will not obtain “beautiful” visualizations but instead a rather good estimate on which buildings can be seen from any viewpoint (Fig. 4

34、(c)). We realized this approach as follows. For any viewpoint, the position and viewing direction define the exterior orientation of a virtual camera of given horizontal and vertical viewing angle. This virtual camera re

35、presents the driver’s view. The height is derived</p><p>　　The virtual image plane is then rastered, each pixel defining a ray in object space. All the rays are traced in object space to determine intersecti

36、ons with the DSM. For each hit, the corresponding object number is obtained by a lookup in an image containing rastered ground plan id’s. Although this method is similar to “ray tracing” used in computer graphics and oft

37、en assumed to be computationally expensive, it is actually quite fast since (a) we are interested only in the first hit of the ray, </p><p>　　2.2 Tracking Visibility</p><p>　　In the last section

38、, visibility was computed for a single view. However, landmarks selected for a routing instruction must be visible during the entire manoeuvre. This can be checked by tracking the visibility of objects along the trajecto

39、ry defined by the corresponding manoeuvre. For our first experiment, we use only a crude approximation for the visibility, namely the area covered by the projection of the corresponding object on the virtual image plane.

40、 </p><p>　　Figure 1 shows an example. We assume that the white polygon is the trajectory we want the driver to use. The question then is if the town hall, identified to be a landmark by the methods of sectio

41、n 5, is a suitable object which can be used in a landmark-based instruction such as ’pass to the right of the town hall’. To this end, our algorithm traces the entire trajectory, generating virtual views at equidistantly

42、 spaced positions and in the orientation de-fined by the trajectory. For each such v</p><p>　　Figure 2 shows a plot of all those areas along the trajectory of figure 1. One can see the typical ’peaked’ curve

43、s generated as objects appear, grow larger and finally disappear as the viewing position passes by. In this special case, one sees also that many objects become visible around frame number 65, which is when the view wide

44、ns as the position leaves the narrow street and enters the plaza in front of the town hall. </p><p>　　In order to answer if the town hall is a suitable object, a look on figure 2 reveals that the correspondi

45、ng curve (shown in bold red) is largest for frame numbers 65 to 115 (with a small exception around frame 100), i.e. the town hall is the largest object in the driver’s view. Moreover, the curve is larger than zero starti

46、ng from frame number 13, which means that the town hall is – at least partly – visible about 100 meters ahead of the position where the plaza is entered (which could be a deci</p><p>　　Figure 1:Example traje

47、ctory, top view.</p><p>　　Figure 2: Visibility plotted over frame number</p><p>　　3 Digital Maps</p><p>　　The maps used by car navigation systems not only contain the geometry and c

48、onnectivity of the road network but also a huge amount of additional information on objects, attributes and relationships. A good overview can be obtained from the European standard GDF, see e.g. (Geographic Data Files 3

49、.0, 1995). Of particular interest are points of interest (POI) which include museums, theaters, cultural centers, city halls, etc.</p><p>　　Map data is acquired by map database vendors such as Tele Atlas or

50、NavTech and supplied to car navigation manufacturers in an exchange format (such as GDF). There, it is converted to the proprietary formats finally found on the map CD or DVD. This conversion is highly nontrivial since t

51、he data has to be transformed from a descriptive form into a specialized form supporting effi-cient queries by the car navigation system. Often, structures and values are precomputed by this conversion process in </p&

52、gt;<p>　　Part of this process is also to generate a matrix for each intersection which describes all possible turn combinations. Also, for the well-known arrow pictograms used by car navigation systems, the angles

53、 between all streets joining at an intersection are stored. </p><p>　　It is during this conversion process where additional information for landmark-based navigation can be integrated. In this paper, we outl

54、ine how the street geometry given by GDF can be combined with information from a cadastral map and laser scan data to identify suitable landmarks. An important point is that the additional datasets are used only during t

55、he conversion process. After that, only landmark-based driving instructions remain, which can be coded in a very compact form and are compatibl</p><p>　　4 LASER SCANNING AND CITY MODELS</p><p>　

56、　During the 1990’s, airborne laser scanning became available as a new method for obtaining surface models. Subsequently, the scanning systems were improved and direct georeferencing became feasible with sufficient accura

57、cy. Today, airborne laser scanning is a mature technology with a multitude of companies offering systems and services (Baltsavias, 1999). Scanning of very large areas is possible, for example the entire Netherlands have

58、been and Germany’s state of Baden-W¨urttemberg is in the prog</p><p>　　The main problem is how to extract symbolic information about man-made structures from laser scanner datasets, possibly combined wi

59、th aerial or terrestrial images. Especially, the automatic generation of city models has been and still is an intense research field, the discussion of which is beyond the scope of this paper. The reader is referred to t

60、he excellent proceedings of the “Ascona workshops” on this topic (Gr¨un et al., 1995, Gr¨un et al., 1997,Baltsavias et al., 2001).</p><p>　　However, there is still substantial research effort neces

61、sary until highly automated object extraction systems working reliably become available. On the other hand, three-dimensional object information is still far from being common in today’s existing GIS databases. In conseq

62、uence, in this paper we consider using two-dimensional GIS databases in combination with laser scanner DSMs on an iconic level, without explicitly reconstructing the three-dimensional shape of the objects as separate ent

63、iti</p><p>　　Figure3:Laser scan</p><p>　　5 CONCLUSION AND OUTLOOK</p><p>　　In this paper, we have outlined how landmarks can be extracted and evaluated using existing GIS and laser

64、scanning data. As for the extraction, we have investigated two different methods based on data mining to reveal prominent buildings. In order to evaluate the usefulness for navigation instructions, we used a visibility a

65、nalysis based on DSM data from laser scanning. </p><p>　　Both data mining procedures have still to be tested with real data sets. The results will verify if they lead to appropriate landmarks in the real wor

66、ld. In addition, the analysis process has to be extended to different object types (traffic constructions, parks, sporting facilities, etc.) extracted for example from ATKIS data. Methods for data preprocessing of differ

67、ent object types and categories, and problems when different data mining algorithms are applied to the same data set, have to be i</p><p>　　More route-dependent aspects to determine real landmarks have to be

68、 investigated: The influence of the users moving direction and visibility on the quality of landmarks. As we only used the “virtual image size” to rate an object’s visibility, there is much room for improvement. For exam

69、ple, from the virtual image, one can also obtain information on the distance, if the object is sticking out behind another, closer object, and if it is part of the silhouette. First pulse laser scan measurements c</p&