201117131 六、發明說明： 【發明所屬之技術領域】 本&明係關於-種用於產生—由複數個小重疊攝影影像 組成之馬赛克的方法，其中歸因於移動前景物體之障礙物 得以最小化。 【先前技術】 數位地圖及數位地圖資料庫用於導航系統中。數位地圖 可藉由各種方法獲得，包括來自空間之高解析度成像以及 自陸用行動車輛獲取之正収影像。在後者情形下，自陸 用製圖系統獲得的影像必須經轉換至一標度經校正且描繪 如^地面特徵之精確地面位置上方觀看之該等地面特徵的 Η正如像。正糾正影像為一種已經幾何校正以使得像片 之軚度均一（意謂像片可被認為等效於地圖）的航拍像片。 彳正如像可用於I測真實距離，因為其為所關注表面 (例如地球表面）之準4表示^針對地形起伏、透鏡崎變 差及相機傾斜而調整正糾正影像。 ° •有效地自航拍景々像獲得正糾正影像。然而，經常引 錯决，此可導致地理定位資料之不準確製圖。一個問題 ^ ^^航拍衫像並非係完全垂直於地球表面而獲取。即使 曰圖像係接近於垂直而獲取時，僅其精叙中（線將為垂 直的。為了正糾正此影像，必須另外獲得地形高度資訊。 "像中物體之準確高度資訊的缺乏結合用於判定正糾 〜像之二角測量方法可導致此等影像之達十幾米或更大 的不正確 。^5TT ii *_ .. 糟由獲取重疊影像並比較自後續影像獲得 144237.doc 201117131 之相同表面而改良正確性。然而，自此方法獲得之正確性 與其成本相比存在限制。 在本文中’術語「水平」資料或資訊對應於具有平行於 或大體上平行於地球表面之表面的物體。術語「垂直」資 料或資訊對應於可藉由大體平行於地球表面之觀視轴而觀 看的物體。垂直資訊不可自典型俯視航拍或衛星影像獲 得。 订動製圖車（通常為陸用車輛，諸如，箱型車或汽車， 但亦可能為航拍車輛）用於收集行動資料以用於增強數位 地圖資料庫。行動製圖車通常安裝有許多相機，可能並令 之一些為立體相機且其全部由於载有精禮Gps及另一位置 及定向判定設備（例如’慣性導航系統携）而經準確地地 理定位。在於道路網或已建立之路線上行驶之同時，以連 續訊框或影像操取地理編碼之影像序列。地理編碼意謂將 由GPS接收器及（可能）INS計算之位置及（可能）與影像相關 聯之額外駛向（heading)及/或定向資料附加至由相機操取 之每-影像的元資料。行動製圖車記錄所關注表面(例 如’路面）之-個以上影像序列，且對於影像序列之每一 影像，準續地判定地理座標參考系統令之地理位置連同該 影像序列相對於該地理位置之位置及定向資料。具有對應 地理位置資訊之影像序列被稱為經地理編碼之影像序列。 其他資料亦可由其他感應器收集’同時且類似地進行地理 編碼。 已知用於獲得正糾正併片⑼e)以用於組合出所關注大表⑴ 244237.doc 201117131 面（諸如，地球）之鳥眼馬賽克（BEM)的先前技術。此技術 之極佳實例描述於申請者之於2008年7月17日公開的國際 公告第WO/2008/044927號中。按認可以引用方式之併入之 權限，特此以引用之方式併入並依賴於該國際公告之全部 揭示内容。 根據已知技術，正糾正影像經組合於一起以產生一馬賽 克而不考慮其中含有之影像内容的品質。實情為，此等影 像通常相繼地依次拼接，與屋頂上木瓦按順序一者重疊於 另一者上極其相似。雖然通常為有效的，但經常發生以下 情形：像片影像中所擷取之移動物體（例如，途經行動製 圖車或由行動製圖車途經之機動車輛）出現於上覆拼片而 非下伏拼片中，使得較不理想之拼片上覆於較理想之拼片 上。因此，部分地使路面之地圖模糊的移動前景物體可能 出現於已完成之BEM中。 申請者之題為「Method Of An Apparatus For Producing A Multi-Viewpoint Panorama」的同在申請中之申請案 P601 5247 PCT描述一種用以使用自行動製圖車之多個觀察 點獲取的垂直影像序列來產生垂直全景之方法。在產生全 景之同時，使用雷射掃描器資料來偵測靠近相機之物體。 藉由標記垂直影像中之不應使用之部分而移除影像中所擷 取的不適宜物體。接著將應使用之部分投射至全景表面 上。 雷射資料之使用（尤其結合垂直影像）為用於產生正糾正 水平影像以用於產生鳥眼馬賽克（BEM)的昂貴、繁瑣且較 144237.doc 201117131 不理想之技術。因此，存在對不依賴於雷 繁殯技術之使用的在所關注表 “或/、他 移動前景物體(尤其當現有影像資::==別 時雷射掃描資料時）的需要。 子在同 【發明内容】 根據本發明，提供—種用於自複數個小重疊 生諸如地球之所關注大表 …象產 自兮斛朗、:t主工 骨兄的方法。该方法包含 自5亥所關注表面之-第—正糾正像片提供—第—拼 驟。s亥弟一像片已藉由—處於一第-距離處之相機碎取^ 該第一拼片經劃分成離散區且與相對於該所關注表面又之。 ::座::置及定向相關聯。針對該第-拼片提=: :值::第，中之所有其他區域指派-低優先二: 離：；弟一光罩經劃分成對應於該第-摒片之該等區之 部分地與該第 之/一正糾正像片提供—至少 一處於第 且之第-拼片。該第二像片已藉由 散==距離處之相機獲取。該第二拼片經割分成離 趵、目、於°亥所關庄表面之-絕對座標位置及定向相 =前:=二拼片產生-第二光罩，其—2 ’7、底兩優先權灰階值且向該第二拼片t之所 八區域指派一低優先權灰階值。該第二 對應於該第二拼片 ，’ w刀成 該m 士 #之離散區。比較該第一拼月與 幵 之—致區（亦即，與相對於該所闕注表面之 相兄絕對座標位置相關聯的區)。本發明之特徵在於·若[S] 144237.doc 201117131 δ亥第一光罩令之該對應區的該灰階值具有一比該第一光罩 之該對應區中之該灰階值高的優先權，則用來自該第二拼 片之該一致區替換該第一拼片令之該一㈣。換言之，藉 由使用辅助光罩標記正糾正影像中之諸部分而移除在馬^ 克中不合需要之移動前景物體。該等辅助光罩允許在兩個 或兩個以上拼片之間缠· #很止 斤乃<間建立優先榷，此使具有最高優先權之 影像能夠使用於馬赛克中，#中具有較低優先權值之影像 被丟棄。因此，可以比使用先前技術可獲得的精度及效率 大的精度及效率產生所關：主女类 干座玍所關主大表面（諸如，地球）之馬赛 克。 【實施方式】 參看諸圖，其中相同數字遍及若干視圖指示相同或對岸 部分，行動製圖車通常以20指示。行動製圖車2〇較佳(但 不必）為安裝有通常使用於地理製圖應用中之類型之一或 多個相機22的陸用笳刑击斗、、卜太 , 一 用相型車或汽車。相機22經高度校準以使 得所關注表面24(諸如，道路）之所獲取圖像可以-特定位 置及定向而進行地理編石馬。此通常係經由自繞地球運行之 複數個衛星28接收位置資料的咖接收器％來實現。此 外’定向判定設備（例如’ INS)係由特徵3〇表示以提供由 相肋獲取之每-影像的驶向資料。藉由此等器件，由相 機22獲取之每一攝影影像經地理編碼，意謂使如由⑽接 _及定向設備30計算出之其位置連同(可能)其他欲向 貝訊與該影像相關聯以作為元資料。當行動製圖車⑽ 路面24時’在時間⑽、，及⑽處擷取路面“之連續影 144237.doc 201117131 像，其中水為連續影像之間的時間間隔。a經建立而足夠 小以使得表面24之連續影像彼此間在區域32處重疊。 如圖2A至圖2C中所展示，多個相機22可結合行動製圖 車20而使用以便在廣泛範圍中且自不同視角記錄表面24之 攝影影像。在對所關注表面24攝影期間，移動前景物體 34(諸如’圖2A至圖2C中所說明之跑車）可能在不同時間相 對於各相機22而暫時阻礙表面24之影像。經模糊化之影像 在其發生於車道合併、交又及其他相關道路特徵上時尤其 令人煩惱，此係歸因於此等特徵在製圖應用中之重要性。 圖3說明行動製圖車2〇遭遇移動前景物體34之另一實 例。在此實例中，面向前及面向後的相機22在不同時間對 同一重疊區域32攝影。重疊區域32僅在一個片刻而非另一 片刻中由移動物體34阻礙.當自複數個小的重疊攝影影像 組合馬赛克（例如，BEM)時，需要使用每一片刻中之最佳 。口質之影像。在對所關注表面24之同一區域32進行一次以 上攝影的情況下（如圖3中）’本發明描述一種可藉以在一影 像内識別出移動前景物體且藉以使更佳品質之影像用於產 生馬赛克的方法。201117131 VI. Description of the invention: [Technical field to which the invention pertains] This & method relates to a method for generating a mosaic composed of a plurality of small overlapping photographic images, wherein the obstacle due to moving the foreground object is minimized Chemical. [Prior Art] A digital map and a digital map database are used in a navigation system. Digital maps are available in a variety of ways, including high-resolution imaging from space and positive-receiving images acquired from land-based mobile vehicles. In the latter case, the image obtained from the land use mapping system must be converted to a scale that is corrected and that depicts the ground features as viewed above the precise ground position of the ground feature. The positive correction image is an aerial image that has been geometrically corrected to make the image uniform (meaning that the image can be considered equivalent to a map).彳 Just as the image can be used to measure the true distance, because it is the quasi-four representation of the surface of interest (such as the surface of the Earth). The corrected positive image is adjusted for terrain fluctuations, lens variability, and camera tilt. ° • Effectively correct the image by taking a picture of the scene. However, this often leads to incorrect decisions, which can lead to inaccurate mapping of geolocation data. A problem ^ ^^ aerial portrait is not obtained completely perpendicular to the surface of the earth. Even if the image is acquired close to vertical, only in its intensive (the line will be vertical. In order to correct this image, terrain height information must be obtained separately. "Lack of accurate height information of objects in combination In the determination of the positive correction ~ image two angle measurement method can cause these images to be more than ten meters or more incorrect. ^ 5TT ii * _ .. bad by obtaining overlapping images and compared from subsequent images to obtain 144237.doc 201117131 The correctness is improved by the same surface. However, the correctness obtained from this method is limited compared to its cost. In this paper, the term 'level' data or information corresponds to an object having a surface parallel or substantially parallel to the surface of the earth. The term "vertical" data or information corresponds to an object that can be viewed by a viewing axis that is generally parallel to the Earth's surface. Vertical information is not available from a typical overhead aerial or satellite imagery. A cartography vehicle (usually a land vehicle, For example, a van or a car, but may also be an aerial vehicle) used to collect action data for enhancing the digital map database. Cars are usually equipped with many cameras, and some may be stereo cameras and all of them are accurately geolocated due to the carrying of the Gps and another position and orientation determination device (such as 'inertial navigation system carrying'). Or a sequence of images that are geocoded in a continuous frame or image while driving on an established route. Geocoding means the location and (possibly) associated with the image that will be calculated by the GPS receiver and (possibly) INS. Heading and/or directional data is attached to the meta-data of each image taken by the camera. The action cart records more than one sequence of images of the surface of interest (eg 'road') and for each of the sequence of images The image is used to determine the geographic location of the geographic coordinates reference system along with the location and orientation data of the image sequence relative to the geographic location. The image sequence with corresponding geographic location information is referred to as a geocoded image sequence. Can be collected by other sensors 'simultaneously and similarly geocoded. Known to obtain positive correction and slice (9)e) ⑴ 244237.doc 201117131 large table surface (such as the Earth) of the bird's eye Mosaic (BEM) in the prior art combination of a concern. An excellent example of this technique is described in International Publication No. WO/2008/044927, filed on Jul. 17, 2008. The right to incorporate the incorporation may be incorporated by reference and is hereby incorporated by reference in its entirety. According to known techniques, positive correction images are combined to produce a Marcel regardless of the quality of the image content contained therein. The truth is that these images are usually stitched one after the other, much like the shingles on the roof overlap one on the other. Although usually effective, it is often the case that moving objects captured in a film image (for example, a motor vehicle passing through an action cart or a motor vehicle passing through a motion cart) appear in the overlying tiles instead of underlying In the film, the less ideal piece is overlaid on the ideal piece. Therefore, moving foreground objects that partially obscure the map of the road surface may appear in the completed BEM. Applicant's application in the same application entitled "Method Of An Apparatus For Producing A Multi-Viewpoint Panorama" P601 5247 PCT describes a vertical image sequence acquired using a plurality of observation points of a self-moving cart The method of vertical panorama. Use laser scanner data to detect objects close to the camera while generating a panoramic view. Unsuitable objects captured in the image are removed by marking portions of the vertical image that should not be used. Then project the part that should be used onto the panoramic surface. The use of laser data (especially in combination with vertical images) is an expensive, cumbersome and less desirable technique for producing a corrected horizontal image for producing bird eye mosaic (BEM). Therefore, there is a need for a watch that is not dependent on the use of the Thunderbolt technique in the table of interest "or /, he moves the foreground object (especially when the existing image material::== when the laser scans the data). SUMMARY OF THE INVENTION According to the present invention, there is provided a method for self-complexing a plurality of small overlaps, such as the large watch of interest of the earth, such as the production of a master, and a master of the body. Focus on the surface - the first - correcting the photo supply - the first - the snippet. The shai ji image has been broken by the camera at a first - distance ^ The first piece is divided into discrete areas and Relative to the surface of interest. :: Block:: Set and orientation related. For the first - piece mention =: : value:: first, all other areas in the assignment - low priority two: away:; brother A reticle is divided into portions corresponding to the regions of the first cymbal and the first/positive corrected image is provided - at least one of the first slabs in the first. The second photo has been borrowed Obtained by the camera at the distance == distance. The second piece is cut into the absolute coordinates of the surface of the 亥 趵, 目, ° And orientation phase = front: = two tiles produce - second mask, - 2 '7, bottom two priority grayscale values and assign a low priority grayscale to the eight regions of the second tile t The second corresponds to the second tile, and the 'w knife is the discrete region of the m. #. Comparing the first spell with the — zone (ie, relative to the surface of the target) a region associated with an absolute coordinate position. The present invention is characterized in that if the [S] 144237.doc 201117131 δ hai first reticle causes the gray level value of the corresponding region to have a ratio of the first mask The priority of the gray level value in the corresponding area is replaced by the same area from the second piece by the uniform area of the second piece. In other words, the image is corrected by using the auxiliary mask mark. Part of the removal removes undesirable foreground objects in the horse. These auxiliary reticles allow for the entanglement between two or more tiles. This enables the image with the highest priority to be used in the mosaic, and images with lower priority values in # are discarded. The accuracy and efficiency of the precision and efficiency that can be obtained by using the prior art are related to the mosaic of the main surface (such as the earth) of the main female class. [Embodiment] Referring to the figures, the same figures The action cart is typically indicated at 20 throughout several views indicating the same or opposite portions. The action cart 2 is preferably (but not necessarily) used for land installation with one or more cameras 22 of the type commonly used in geographic mapping applications. A smashing, squad, a phase car or a car. The camera 22 is highly calibrated so that the acquired image of the surface 24 of interest (such as a road) can be geo-arranged in a specific location and orientation. This is typically accomplished by the percentage of coffee receivers that receive location data from a plurality of satellites 28 operating around the earth. Further, the orientation determining device (e.g., 'INS) is represented by feature 3〇 to provide heading data for each image acquired by the phase ribs. With such a device, each photographic image acquired by camera 22 is geocoded, meaning that its position as calculated by (10) and orientation device 30, along with (possibly) other desires to associate with the image. Take as metadata. When the action cart (10) is on the road at 24 o'clock, at time (10), and (10), draw a continuous image of the road surface 144237.doc 201117131, where water is the time interval between successive images. a is established to be small enough to make the surface The successive images of 24 overlap each other at region 32. As shown in Figures 2A-2C, a plurality of cameras 22 can be used in conjunction with the mobile cart 20 to record photographic images of the surface 24 over a wide range and from different viewing angles. During photography of the surface 24 of interest, moving the foreground object 34 (such as the sports car illustrated in Figures 2A-2C) may temporarily obstruct the image of the surface 24 relative to each camera 22 at different times. The blurred image is This is particularly annoying when it comes to lane consolidation, intersections, and other related road features, due to the importance of these features in mapping applications. Figure 3 illustrates that the action cart 2 encounters moving foreground objects 34 Another example. In this example, the front-facing and rear-facing cameras 22 photograph the same overlapping area 32 at different times. The overlapping area 32 is moved only one moment instead of another. The object 34 obstructs. When combining a small number of small overlapping photographic images with a mosaic (for example, BEM), it is necessary to use the best image of each of the moments. The image of the same area 32 of the surface 24 of interest is performed more than once. In the case of photography (as in Figure 3), the present invention describes a method by which a moving foreground object can be identified within an image and whereby a better quality image is used to create a mosaic.
的—者進行正糾 且在此特定例子 圖5展示在已使用上文所描述之技術中的— 正之後的像片。正糾正影像被稱為拼片，且 144237.doc -9- 201117131 拼片36，但彼術語為稍微隨意的 中被稱為「第二 此，對於任何：定時刻…的。因 尜入夕从 T應於？之正糾正影像根據其中 肷入之地理編碼資料而置於 拼片)及_三拼片統巾，於 座標系統中，因此可發現^ 影像經置於相同 衫像之間的重疊部分。已在^ 中以圖形方式描繪此。 =特定參看圖5,在此階段不知曉第二拼片％ :區^移動物體及影像之哪—部分與所關注表面“有 界“楚起見’此處之術語「區」用於描述整個拼片之 二域“务上，將針對數位像片中之每一像 = 而’達到彼精細標度之解析度並非始終為 展示圖5之第二拼片娜「第一」拼片_ 重疊口Ρ分32。第一拼只38矣-丄i 分 弟拼片38表不由相機22在時間卜」f處或緊 接於獲取導致正糾正第-拼 A緊 拚片36之攝影影像的時間之前獲 取的攝:影像。如隨後將描述’有可能第一拼片％及第二 开片6由兩個不同相機22同時獲取或由兩個不同㈣在兩 個不同時間處獲取（如圖3中所提示）。 非=片Μ以圖6中所展示之方式重疊時，所關注之 非移動表面24顯得大體上相同’使得影像可經重疊而具有 極少至無失真。此可由重疊區域32中之完全對準之車道桿 ㈣實1而’移動物體34在時間以與時間遠具有不同 位置’且因此可在沿道路之正糾正拼片％、％之重疊部分 —在/σ路面24之不同位置處看到。重疊區域η亦可被稱為 —致區32’意謂拼片36、38中之此等各別區（或像幻與相 144237.doc 201117131 對於所關注表面2 4之相同絕對座標位置相_。 圖7及_別描繪第二拼片36及第—拼片取一致區 二亦即’圓7為第二拼片36之片段視圖，其僅展示其一 分。另一方面’圖8為第一拼片％之片段視圖， :、展不，、相同一致區32部分。在比較第一拼片财第二拼 36之一致區32的過程中，顯然在圖7中未阻礙路面24， 而在圖8中路面之部分由移動物體34阻礙。藉由比較拼片 %、36之重疊部分’有可能判定是否存在運動中之物體 此_由逐區域或逐像素地計算灰階值之絕對差而進 仃。接著對此等進行定限以獲得被稱為如圖9中所描緣之 光罩40的黑色/白色影像。無論係逐像素還是以較粗略之 區域分析進行，灰階值均係跨越整個一致_針對第一拼 片38及第二拼片36中之每一者而判定。The person performing the corrective and in this particular example Figure 5 shows the picture after the use of the technique described above. The corrective image is called a patch, and 144237.doc -9- 201117131 is a piece 36, but the term is slightly random. It is called "the second one, for any: fixed time... because of the eve T should correct the image according to the geocoded data inserted therein and place it in the piece) and _ three pieces of the film in the coordinate system, so it can be found that the image is placed between the same shirt image. Part. This has been graphically depicted in ^. =Specific reference to Figure 5, at this stage, the second piece % is not known: the area ^ the moving object and the image - part and the surface of interest are "bounded" 'The term "zone" is used to describe the second domain of the entire tile. "Of course, it will be for each image in the digital image = and the resolution of the fine scale is not always shown." Two pieces of the film "first" piece _ overlapping mouth points 32. The first spell is only 38矣-丄i. The splitter 38 is not taken by the camera 22 at the time or immediately before the time at which the photographic image of the first stitch A is being corrected. image. As will be described later, it is possible that the first tile % and the second opening 6 are simultaneously acquired by two different cameras 22 or acquired by two different (four) at two different times (as prompted in Fig. 3). When non-slices overlap in the manner shown in Figure 6, the non-moving surfaces 24 of interest appear substantially the same' such that the images can be overlapped with little to no distortion. This can be done by the fully aligned lane bar (4) in the overlap region 32 and the 'moving object 34 at a different position from time to time' and thus can correct the overlap of the pieces %, % along the road positively - at /σ Road 24 is seen at different locations. The overlap region η may also be referred to as the region 32', meaning that the respective regions in the tiles 36, 38 (or the same absolute coordinate position for the surface of interest 144237.doc 201117131) Figure 7 and _ do not depict the second tile 36 and the first tile take the same area 2, that is, the circle 7 is the segment view of the second tile 36, which only shows one point. On the other hand, Figure 8 The fragment view of the first piece %, :, the exhibition is not, the same consistent area 32. In the process of comparing the consistent area 32 of the first piece of the second piece 36, it is apparent that the road surface 24 is not obstructed in FIG. In Fig. 8, the portion of the road surface is obstructed by the moving object 34. By comparing the overlapping portions of the tiles %, 36, it is possible to determine whether there is an object in motion. This is calculated by region-by-region or pixel-by-pixel. This is followed by a limit to obtain a black/white image called a reticle 40 as depicted in Figure 9. Grayscale values, whether pixel-by-pixel or coarser-area analysis All are determined across the entire consistency _ for each of the first tile 38 and the second tile 36.
灰階值通常處於〇至255之範圍中，其十〇相當於黑色且 55相田於白色。在衫色像片中，可籍由針對每—區或像 素簡單地平均個別紅色、綠色及藍色色值而計算灰階值。 因此’根據簡單平均技術’若紅色色值為155、藍色色值 為且”、彔色色值為90，則灰階色值為約86。然而，實務 上，灰階值通常計算為加權總和。舉例而言·· 〇.2989xR+0.587〇xG+〇 114〇χΒ。當然，亦可使用其他灰階 判定技術。適當臨限值經預定而處於數字〇與255之間。舉 例而。臨限值可選擇為60。在此情形下，若第一拼片38 第拼片36中之一致區之像素或區域中的灰階值之間的 、巴對差（亦即，差之絕對值）超過臨限值（例如，00)，則移[S 144237.doc 201117131 動前景物體34經識別為存在於彼像素或區域中。作為一實 例’若第一拼片38中之一致區32内之特定像素或區域的灰 P白值為86且第—抑片36之對應像素或區域中的灰階值為 1_5，則灰階值之間的絕對差等於86減15或7卜差71高於例 不性臨限值60，且因此斷定移動前景物體34經描繪或摘取 於一致區32之彼特定像素或區域中。 藉由以此方式比較兩個拼片36、38，可產生可被稱為第 光罩40(因為其與第—拼片％相關聯）之光罩扣。當第一 拚片38與第一抑片36之間的灰階值中之絕對差低於預定臨 限值時’第-光罩4〇將白色灰階值（亦即，255)指派給第一 光罩40中之對應像素或區域。然而，當絕對差之計算產生 高於預定臨限值之數字以使得移動前景物_經識別為存 在於第二拼片36之彼像素或區域中時，向光罩40之對應像 素或區域指派-黑色灰階值（亦即，〇)，如圖9中之黑色區 域所表不。因此，在上文所提及之實例（其中灰階值之絕 、為）中’光罩4G中之彼特定像素或區域將經指派一 黑色灰階值或呈現黑色’如圖9中所展示。藉由此方法， 光罩清楚地識別其㈣動前景物體％之像素或區 域。 :‘、田兩個對應像素（或區域）之間的絕對差超過臨限 值可易於藉由將255而非〇指派給一像素而顛倒此等 白色」及「黑色」慣例…種用以解釋本發明之此特徵 之完全不同的方式避免術語「白色」及「黑色」之潛在複 雜之使用’且替代地僅關注像素優先權或重要性。在此情 144237.doc 201117131 形下’可在灰階值比較之基礎上嚴格地評估像素（或區域） 優先權。在臨限值設定（在先前實例中，僅出於論述目的 而提示為「60」）之側上的絕對差比較被給予高於落於臨 限值之相對侧上之彼等比較的優先權。因此，在—方法 中’較低值（亦即，低於臨限值）意謂較重要之像素，而在 另一方法中，較高值意謂較重要之像素。此僅為用以解釋 光罩值之使用及實施的另一方式。 或者，並非向光罩40之對應像素或區域指派黑色〇(戋白 色255)灰階值，可能較佳將某一中間灰階值指派給光罩⑽ 中之對應像素或區域，該中間灰階值可能等於第—拼片38 之一致區32中所計算之灰階值。換言之，若第—拼片“中 之一致區32中的對應像素或區域具有灰階值71，且絕對差 之計算超過預定臨限值，則光罩4〇中之對應區域或像素將 被給予一中間灰階值71。此為上文所描述並展示於圓^中 之方法的替代方法，使得光罩4〇將顯示介於臨限值（例 如，60)與0(或在白色黑色慣例如先前所描述而顛倒時為 255)之間的灰階值。在任何情形下，值得注意的係：光罩 4〇係藉由兩個拼片36、38之比較而產生，其中移動前景物 體34係藉由計算一致區32中之對應像素或區域之灰階值的 絕對差而識別。 圖1 〇提供使用（諸如）可用於使用以賦能軟體（enabling software)程式化之電腦處理器之實際應用中的功能模組之 方法步驟之概述。以簡化方式展示一用於使用光罩產生馬 賽克之方法流程。根據此技術，收集道路24之正糾正影[s] 144237.doc •13- 201117131 像，該等正糾正影像已藉由安裝於行動製圖車2〇上之經护 準視覺設備22而記錄。每一影像嵌入有與正糾正拼片對^ 之位置資料。接著藉由比較重疊拼片而產生光罩，藉吨 供關於正糾正拼片中之一致區32的每一區域或像素之品質 的資訊。接著’可使用此等光罩來產生極大所關注= 24(諸如，地球之表面）之馬賽克。 以此方式，藉由比較重疊正糾正影像而針對每—正糾正 拼片產生光罩。然而，如下文中較全面地描述，可使用特 定模型化或預測技術來預測移動物體34何時將處於特定拼 片影像中且接著僅針對彼等拼片產生光罩。可藉由比較光 罩之序列而增強或改進移動物體34之偵測，如可能最佳展 示於圖U至圖"中。舉例而言，圖⑴苗繪如圖5甲所展示 之正糾正第二拼片36。為了改良原始偵測結果，可模型化 移動物體34之行為。移動物體34通常屬於兩種類別：相對 於行動製圖車2〇處於大體上恆定速度之物體，及正追上行 動製圖車20或正被行動製圖車2〇追上之移動物體34。雖然 在行動製圖車20前方行進之第一類別中的物體叫實變得 在拼接影像之頂部部分中可見，但其亦自該影像之相同部 分中消失不見。此等物體34在實際處理上並不困難，因為 當連續拼片一者重疊於另一者上以產生所得馬赛克時，其 被「拼除（tile away)」且在最終馬赛克中幾乎始終不可 見，此係因為不含有該物體之下一拼片經繪製於其上，與 屋頂瓦片極其相似。因此’第二類別（追上）中之物體湘 向於引起較大困難。此等物體34傾向於出現於所得馬賽克 144237.doc 201117131 :EM中)或拼片中且幾乎始終行敬於與行動製圖車20不同之 車：中，此係歸因於追趕汽車之特定性 以用於說明）。 固i 同=繪展示在時間t，在正糾正影像或拼片之四個不 B、C、咐之光罩資㈣原始偵測資料。因 二圖U中之數字卜2、3·.·表示關於特 上之區心 編號。根據圖U，垂直轴表示水平方向 立^物之光罩。此處，影像之頂部部分+的黑色 :=ΓΓ，片之左邊(見圖")。在影像之底部 物體’’、、色思明在光罩產生之第-步驟(圖6)期間偵測到 ==、’此偵測到原始移動障礙34。因此’在特定地參看 "12Β及圖13時，水平訊框經劃分成四個垂直區 個區域為全部黑色或全部白色。在於完成原 早、測之後具有特定值之彼區域（八至咐，藉由對像 數定限而選擇值(用於黑色之。及用於白色之255)。 Μ ’為了改Ml㈣’基於移動通過訊框之物體34的經 杈5L化仃為而調整資料。結果為如圖中所說明之資 料圖12B隨時間推移較清楚地描績在前^個訊框中相當 快速地追上行動製圖車2()之物體34及接著在訊框Η至約5〇 中追上行動製圖車20的慢得多之移動物體34。接下來的10 個訊框(大約)不含有所偵測移動物體，然而，訊框70至 85(大約）展示追上移動物體34之行動製圖車2〇。 S] 每-光罩可描述為指示正糾正影像(亦即，拼片)中之哪 些區域或像素含有運動中之物體34的資料集合。由圖12A[ H4237.doc -15- 201117131 及圖12B說明之治& 則述實例描述偵測資料之改進以產生 結果。此等步驟 座生更佳 非針對視覺系統之每一組件而執行，而 系僅針對特定子集而執 為先前步驟之輸出而^ 料易於作 為可用的。然而，基於彼子集的摘測 、，、。果及關於行動製圖車2〇之視覺系統之設置的知識，亦可 針對視I系統之所有組件之每—正糾正影像而產生光罩資 料2本原理為記錄視覺系統之不同組件（包括相機Μ)不 同女裝於仃動製圖車2。上。此意謂在時間t處，路面Μ 上之物體34可在由視覺系統之不同組件記錄的多個垂直^ 像中之不同位置處看到。因此，給㈣於視覺系統之至；; 組件22運動中之物體34在路面24上之位置及移動的知 識’可預測路面24上之移動物體34將在視覺系統之里他么且 件的影像中處於何處及是否可見，且亦可針對彼等組件產 生光罩資料。 作為只例，相機22之子集可為兩個側部相機（左/右）， 且光罩係藉由在僅針對彼兩個相機之正糾正空間中的差分 而產生。基於此等結果’假定移動物體34符合以下假設， 則可針對其他相機（例#，前部相機及後部相機）產生光 罩：對於視覺系統之每-組件，若運動中之物體在時間" 及在時心2處在正糾正影像中可見，則預期其對於所有，亦 σ見’、中"< ^ < G ’且預期在時間"處在正糾正影像之 -部分中變得可見的遠多於—個之物體在時間。處在影像 之相對部分中移離可見度（_e⑽Gf仏⑻㈣。因此， 在右側相機22上變得可見之物體34針對右前方相機產生光 144237.doc 201117131 罩以使得使用此一者。由於視角差異，道路24之在側部相 機22中被阻擋之部分在前部相機22中仍可見，因此可使用 來自前部相機之影像。一旦追趕汽車變得亦可見於右側相 機之左邊中且右邊部分再次變得不可使用，便可針對前部 相機產生光罩以使得在此情形下不使用彼一者（因為障礙 34將愈加可見）。因為每一相機22之駛向及子集申之相機 的駛向為已知的且僅基於子集相機之光罩中的彼角度，所 以亦針對其他相機產生光罩。只要正糾正空間中之訊框之 間的共同部分足夠大’便可能明確地針對每一相機產生光 罩。然而，僅使用一精選子集大大增加處理速度且僅稍微 降低結果❶因此，障礙之行為愈符合上文陳述之假設，則 注意到的效能之降低愈小。 如上文所陳述，光罩可被解釋為加權影像。黑色（亦 即，灰階值255)意謂最低優先權，而白色意謂最高優先 權。光罩產生方法流程中之最初兩個步驟僅產生黑色或白 色值。如先前所提示，第三步驟可產生小於乃5之灰階 值，藉此基於子集相機之光罩及相機之角度而將不同優先 權給予不同相機。 藉由此等方法’有可能最佳化自垂直影像得到之正糾正 拼片36、38的產生以便改良路面及路肩之可見度。因為所 關注表面24上之相同點可在相同時間或不同時間自兩個不 同相機22(或在不同時間自相同相機22)可見，所以可使用 本發明之概念實現改良之可見度。 圖14說明此技術之另—流程圖’其中讀取自第―正糾正 144237.doc •17- 201117131 像片提供之第—描 片（步驟42)連同第一拼片之第— 驟44)，此識別第—拼 先罩（步 於論述目的，可卜/中之任何已知移動前景物體。出 面之馬搴吞m 片連同其第一光罩構成地球表 奢/ 如，BEM)的現有部分。在步驟46中，提供 ^ 了部分地與第一拼片重疊之新正 = 二=統讀取其位置資料。將第二光罩投射至臨= 二所指示。同樣，在步驟%中，將第二二 X H時光料片上。在步驟52處，計算臨時拼 域所旦機距：4此為自相機22之焦點至考慮中之像素或區 '里測的歐幾里德距離。逐區域或可能逐像素地比較 一拼片與第二拼片之全部一致區32。若第一或目的拼片且 有空區域或像素，則使用來自第二、臨時拼片之對應區域 或像素。此展示於詢問54及步驟56中。若第二（亦即，臨 時）光罩中之對應像素或區域的灰階值大於第一光罩中之 對應像素或區域的灰階值，則用來自第二或臨時拼片之像 素或區域替換第一拼片中之彼像素或區域。此展示於步驟 56之前的詢問58中。若灰階值相等或處於預定義範圍之内 (如詢問60所提示）’則在62處作出另一詢問以判定第二、 臨時像素之相機距離是否小於第一、目的像素之相機距 離。若第二、臨時像素係自較近距離獲取，則根據步驟 5 6，將第二、臨時像素（或區域）複製至（亦即，替換）第一 像素（或區域）。接著更新光罩值（步驟64)以及相機距離（步 驟66中）。在68處作出關於是否已考慮一致區中之最末區 域或像素的詢問。若否’則重複方法步驟52至66。一旦已 144237.doc 18- 201117131 以此方式分析來自一致區之最末像素（或區域），便在步驟 70中保存經更新拼片連同經更新光罩，且該經更新拼片及 該經更新光罩變成馬賽克（BEM)之部分。 參看圖15至圖18C，以圖形方式表示圖14之圖。在此等 貫例中，第一拼片38由指向前之相機22表示且第二拼片36 來源於成角相機22。然而，必須理解，展示於圖15中之相 機22之特定定向嚴格地僅用於說明目的。正糾正第一拼片 38展示於圖16A中，而正糾正拼片“展示於圖口八中。針 對第一拼片38產生之光罩40展示於圖16B中，而第二拼片 36之光罩72展示於圖17B中。在此簡化實例中，僅在第二 拼片36中偵測到移動物體34(圖丨7A)，其對應光罩μ反电 其中之所識別移動物體34 m光罩影像兩者較佳儲名 於例槽案中。如圖16B中所展示，在圖似之拼片％中不 存在待遮蔽之物’因為在水平影像中尚未偵測到移動杉 體。因此’光罩40完全為白色。第二拼片财其光罩^展 不於圖Μ及圖17B中。接著，拼片36、38如圖i8a中所展 :而重疊而無光罩’使得當第二拼片％上覆第一拼片” 時，移動物體34使圖16A中清楚看到的路面影像之部分模 糊。接著，在圖刚中將光罩4〇、72展示為經組合。若藉 由比較第-拼片38及第二耕片36中之一致（亦即，重疊）區 ”，第二光罩72令之灰階值大於第一光罩4〇中之灰階值， 則將使用來自第二拼片36之—致區來替換第-拼片38之一 ^區。^’在此特定實例中為相反情形，因為在兩個光 4〇、72中對應之—致區的比較展示第二光罩72中之灰階 144237.doc •19· 201117131 r :於f光罩40中之對應區的灰階值。因此，使用第一 开38影像中之下伏部分，如由所得圖18C所表示。 克中圖動物體34之某一部分確實出現於㈣ 之景m ^片36含有在第—拼片38中不存在對應區 區域碎貝… 當在第—拼片38中不存在對應像素或 ^時’使用來自第二拼片36之影像資料，即使其含有已 之°=:!34。在—致區之比較指示第-光罩與第二光罩 y =階值大體相等的情形下，則系統將評估獲取各別 及二像片之距離。此處之攝影距離表示正糾正 像與相機22之焦點之間的距離。具有最小攝影距離 給認為較可靠’且因此其影像將在重疊區η中被 -旦重疊完成’便更新馬賽克光罩連同記錄於馬賽克中 之像片距離’以使得在任何後續拼接操作巾，新正糾正拼 片將與所記錄光罩資料進行時1此方式將正糾正拼片 組合成馬賽克，其中重疊區係基於特定地關於移動物㈣ 之存在的影像内容而選擇。 由本發明之技術，識別出移動物體3 4且接著自 正糾正拼片產生光罩，此可用於在產生所關注大表面 24(諸如，地球）之馬賽克時判定重疊拼片之哪些區應被給 予優先權。根據先前技術，不加選擇地上覆之正糾正拼片 Τ、.’α出不太有用之結杲，因為障礙34可覆蓋所關注表面2斗 之部分。然而，根據本發明，光罩之使用幫助選擇具有水 平物體（諸如，分道線、車道走廊、排水溝置放等）之最相 144237.doc • 20- 201117131 關資訊的最佳可用影像。因此，光罩之使用幫助改良所得 馬赛克（BEM)之可辨性。且，因為此等光罩可嚴格地在所 比較影像資料之基礎上產生，所以不需要額外成像或雷射 資料技術來識別移動物體34。實情為，僅需要一對重疊水 平（正糾正）影像來產生鳥眼馬賽克（BEM)。藉由多個正糾 正拼片之共同區域或像素的灰階差分而偵測移動物體h。 與在垂直訊框情形下改變偵測相反，因為偵測係在正糾正 空間中進行’所以該方法可直接區別背景與移動物體 圖19展示本發明之兩個替代應用’其中正糾正拼片係由 自載運於航拍車輛（諸如，衛星12〇或飛機22〇)上之相機 122、222所獲取的影像而產生。又，在此情形下，移動前 景障礙物134、234可在所得影像中產生障礙物。經由本文 所描，之概念的直接應用’有可能改良來自此等航拍影像 之所得馬賽克的影像品質。 根據相關法定標準對前述發明進行了描述，因此該摇 述性質上為例示性的而非限懸的。對所揭示實施例^ =改對於熟習此項技術者而言可變得顯而易見且落入 笳……山 #本發明所賦予之法定保護之 【圖式簡單說明】 辄可僅可錯由研究以下中請專利範圍而判定。 圖為穿越路面且使用適當攝影設備獲取一系列順序參 像之行動製圖車的高度簡化說明，該系列順序 ： GPS定位資料連同自適 地理編碼； 冑取付之疋向貧料而進行 S] 144237.doc 201117131 圖2A至圖2C說明一時間順序視圖，其中根據本發明之 行動製圖車被一前景移動物體（在此情形下描繪為跑車）追 上； 圖3為展示跟隨一移動前景障礙物之行動製圖車的時間 推移序列’該移動前景障礙物部分地使藉由—個（面向前 的）相機獲取之所要路面的影像模糊但並未使藉由不同（面 向後的）相機獲取之相同表面的影像模糊； 圖4為如自行動製圖車（諸如，圖3中所描繪之行動製圖 車）頂部之面向前的相機觀看的簡化透視圖，其中移動前 厅、P早礙出現於前方左車道中，且靡魂矣千士上 且嚴踝表不由刖向相機獲取 之攝影影像的邊界； 圖5表示圖4之攝影影像之正糾正視圖，其中展示為左 角中之變黑部分的障礙使路面之視圖模糊； w 圖6為圖5中所描繪之拼片連同先前第一拼片㈣的」 覆物，其經配置以展示移動前景障礙物可自一個拼片至] 一拼片改變相對位置且可在一個拼 、， ^個拚片中產生視圖障礙物汗 在另一拼片中不產生視圖障礙物的方式； 圖7描繪如圖5中所展示之第二拼片之一致區； 圖8為圖6之第一拼片之—致 A •的視圖，其中移動障礙形 經展不而阻擋路面之一部分； 圖9描繪用於第—拼片之一致區（圖8)的光罩； 圖1〇為描述使用本發明之方法產生馬赛克的流程圖； 圖11表示類似於圖5之τ ω 之正糾正拚片之正糾正拼片，1屮 於處理後影像改進之目的而$〜 〈曰的而經再劃分成四行（八至…； 144237.doc -22- 201117131 圖12A為自本發明收集之原始資料的時間圖，其中列表 示每一拼片中之經再劃分之區（八至D)且行表示順序拼片或 影像 〇△/、ί、ί + &等）； 圖12Β為圖12Α之時間圖，其說明可使用行為模型化來 改良前景移動物體之偵測的方式； 圖13為圖12Α中之侷限於13處之區域的放大視圖； 圖14為描繪用於使用光罩改良拼接之沿道路之正糾正影 像中之路面之可見度的步驟序列之流程圖； 圖15為安裝有複數個相機之行動製圖車的簡化俯視圖， 兩個此等相機同時對所關注表面上之重疊區域攝影； 圖16Α描繪由圖15中之行動製圖車之指向前的第一相機 所擷取的第一拼片； 圖16Β為針對圖16Α之第一拼片而產生之光罩； 圖17Α為自圖15中之行動製圖車成角地面向之第二相本 所獲取的正糾正第二拼月； ’ 圖17Β表示針對圖17Α之第二拼片而產生之第二光罩； 圖18Α表示第一拼片及第二拼片之拼接，其中重疊之筹 二拼片歸因於移動前景障礙物而使可見路面之部分模糊； 圖⑽描緣第一光罩與第二光罩之間的比較，其中光罩 優先權經評估且用於判定第1片及第二拼片之哪些部分 含有所關注表面之較準確資料； 圖18C為如圖IgA中之視圖，妙 ^ s ϋ然而，圖18C展示使用藉由 先罩比較所得之改良資料來產生馬赛克； 圖19為纟兄明本發明之槪令 S] Μ以心其㈣像獲取及馬賽克 144237.doc -23- 201117131 應用的方式之高度簡化視圖，其中正糾正拼片可來源於衛 星影像及/或航拍像片。 【主要元件符號說明】 13 區域 20 行動製圖車 22 相機 24 路面 26 GPS接收器 28 衛星 30 定向設備 32 重疊區域 34 移動前景物體 36 第二拼片 38 第一拼片 40 光罩 72 光罩 120 衛星 122 相機 134 移動前景障礙物 220 飛機 222 相機 234 移動前景障礙物 A 區域 B 區域 144237.doc -24- 201117131 c 區域 D 區域 144237.doc •25-The grayscale value is usually in the range of 〇 to 255, with ten turns equivalent to black and 55 phased to white. In the shirt color image, the gray scale value can be calculated by simply averaging the individual red, green, and blue color values for each zone or pixel. Therefore, according to the simple averaging technique, if the red color value is 155, the blue color value is "," and the color color value is 90, the gray color value is about 86. However, in practice, the grayscale value is usually calculated as a weighted sum. For example, 〇.2989xR+0.587〇xG+〇114〇χΒ. Of course, other grayscale determination techniques can be used. The appropriate threshold is predetermined between the numbers 255 and 255. For example, the threshold can be The selection is 60. In this case, if there is a bar-to-difference (i.e., the absolute value of the difference) between the grayscale values in the pixels or regions of the uniform region in the first tile 38, the patch 36 exceeds The limit value (eg, 00) is then shifted [S 144237.doc 201117131 The moving foreground object 34 is identified as being present in the pixel or region. As an example, if a particular pixel in the uniform region 32 in the first tile 38 is present Or the gray P white value of the region is 86 and the grayscale value in the corresponding pixel or region of the first film 36 is 1_5, then the absolute difference between the grayscale values is equal to 86 minus 15 or 7 dip 71 is higher than the case The threshold value 60, and thus concludes that the moving foreground object 34 is depicted or extracted from the specific image of the consistent region 32. Or in the region. By comparing the two tiles 36, 38 in this manner, a hood buckle that can be referred to as a reticle 40 (as it is associated with the first tile %) can be created. When the absolute difference among the gray scale values between 38 and the first sheet 36 is lower than the predetermined threshold, the first photomask 4 指派 assigns a white gray scale value (ie, 255) to the first mask 40. Corresponding pixel or region. However, when the calculation of the absolute difference produces a number above a predetermined threshold such that the moving foreground is identified as being present in the pixel or region of the second tile 36, the reticle 40 The corresponding pixel or region assignment-black grayscale value (ie, 〇) is represented by the black region in Fig. 9. Therefore, in the above-mentioned example (where the grayscale value is absolute) 'The particular pixel or region of the reticle 4G will be assigned a black grayscale value or appear black' as shown in Figure 9. By this method, the reticle clearly identifies the pixel or region of the (4) moving foreground object. : ', the absolute difference between the two corresponding pixels (or regions) exceeds the threshold can be easily 255 instead Assigning to a pixel and reversing such white and black conventions... a completely different way of explaining this feature of the invention avoids the potentially complicated use of the terms "white" and "black" and instead focuses only on Pixel priority or importance. In this case, the pixel (or region) priority can be strictly evaluated based on the grayscale value comparison. The absolute difference comparison on the side of the threshold setting (in the previous example, only "60" for discussion purposes) is given priority over the comparisons on the opposite side of the threshold. . Thus, in the method, the lower value (i.e., below the threshold) means the more important pixel, and in the other method, the higher value means the more important pixel. This is just another way to explain the use and implementation of the mask values. Alternatively, instead of assigning a black 〇 (戋 white 255) grayscale value to a corresponding pixel or region of the reticle 40, it may be preferable to assign an intermediate grayscale value to a corresponding pixel or region in the reticle (10), the intermediate grayscale The value may be equal to the grayscale value calculated in the coincident region 32 of the first tile 38. In other words, if the corresponding pixel or region in the uniform region 32 of the first tile has a grayscale value of 71, and the calculation of the absolute difference exceeds the predetermined threshold, the corresponding region or pixel in the mask 4〇 will be given An intermediate grayscale value of 71. This is an alternative to the method described above and shown in the circle, such that the reticle 4〇 will display between a threshold (eg, 60) and 0 (or a white convention in white) Grayscale values between 255) when inverted as described previously. In any case, it is worth noting that the reticle 4 is produced by comparison of two tiles 36, 38, wherein the foreground object is moved The 34 is identified by calculating the absolute difference of the grayscale values of the corresponding pixels or regions in the coincidence region 32. Figure 1 provides for use with, for example, a computer processor that can be programmed with an enabling software. An overview of the method steps of the functional modules in practical applications. A method flow for generating mosaics using a reticle is shown in a simplified manner. According to this technique, the positive correction of the road 24 is collected [s] 144237.doc •13- 201117131 Like, these are correcting the shadow It has been recorded by the compliant visual device 22 mounted on the action cart 2. Each image is embedded with the position data of the positive correction patch pair. Then the mask is produced by comparing the overlapping tiles. Tons of information about the quality of each region or pixel that is correcting the uniform region 32 in the tile. Then, use these masks to create a mosaic of great attention = 24 (such as the surface of the earth). In this manner, a reticle is created for each positive correction tile by comparing overlapping positive correction images. However, as described more fully below, specific modeling or prediction techniques can be used to predict when a moving object 34 will be in a particular tile image. The reticle is then generated only for the tiles. The detection of the moving object 34 can be enhanced or improved by comparing the sequences of the reticle, as best shown in Figures U through. For example, Figure (1) Miao painted as shown in Figure 5A is correcting the second tile 36. To improve the original detection results, the behavior of the moving object 34 can be modeled. The moving object 34 generally belongs to two categories: relative to the action system The vehicle 2 is at a substantially constant speed and is catching up with the action cart 20 or the moving object 34 being chased by the action cart 2, although the object in the first category traveling in front of the action cart 20 is called It becomes visible in the top portion of the stitched image, but it also disappears from the same portion of the image. These objects 34 are not difficult to handle in practice because one of the consecutive tiles overlaps the other. When the resulting mosaic is produced, it is "tile away" and is almost invisible in the final mosaic, because it does not contain a piece of the object below it, which is drawn on it, much like a roof tile. . Therefore, the object in the second category (catch up) tends to cause greater difficulty. These objects 34 tend to appear in the resulting mosaic 144237.doc 201117131 : EM) or in the tiles and almost always respect the car different from the mobile cart 20: this is due to the specificity of catching up with the car. Used for explanation). Solid i Same as = painted at time t, in the correction of the image or the four of the tiles are not B, C, 光 光 ( ( (4) original detection data. Because the number 2, 3·.· in the figure U shows the number of the zone heart. According to Figure U, the vertical axis represents the reticle of the horizontal direction. Here, the black of the top part of the image is +=ΓΓ, to the left of the slice (see figure "). At the bottom of the image, the object '', and the colorimetric detection during the first step of the mask generation (Fig. 6) ==, 'this detects the original movement obstacle 34. Therefore, when specifically referring to "12" and Fig. 13, the horizontal frame is divided into four vertical areas, all black or all white. It is to complete the original area and the area with a specific value after the test (eight to 咐, select the value by the limit of the number of images (for black. and 255 for white). Μ 'To change Ml (four)' based on mobile The data is adjusted by the 杈 5L of the object 34 of the frame. The result is the data as illustrated in the figure. Figure 12B clearly clears the action charting in the first frame in time. The object 34 of the car 2 () and then the target frame car 追 to about 5 追 catch up with the much slower moving object 34 of the action cart 20. The next 10 frames (approximately) do not contain detected moving objects However, frames 70 through 85 (approximately) show an action cart 2 that catches up with the moving object 34. S] Each reticle can be described as indicating which regions or pixels in the image (i.e., the tile) are being corrected. A collection of data containing objects 34 in motion. The example described in Figure 12A [H4237.doc -15-201117131 and Figure 12B illustrates the improvement of the detection data to produce a result. These steps are better. Executed for each component of the vision system, but only for a specific subset The output of the previous steps is easy to use. However, based on the subset of the measurements, the results, and the knowledge of the settings of the visual system of the mobile cart, you can also target all components of the I system. Each—correcting the image to produce the reticle data 2 The principle is to record different components of the vision system (including the camera Μ) different women's clothing in the drawing cart 2. This means that at time t, the road surface is on The object 34 can be seen at different locations in the plurality of vertical images recorded by different components of the vision system. Thus, (4) to the vision system;; the position of the object 34 in the motion of the assembly 22 on the road surface 24 and The knowledge of movement 'predicts where the moving object 34 on the road surface 24 will be in the image of the visual system and in the image of the piece, and can also produce reticle data for its components. As a case, the camera The subset of 22 can be two side cameras (left/right), and the reticle is produced by the difference in the positive correction space for only the two cameras. Based on these results, it is assumed that the moving object 34 meets the following Hypothesis A mask can be created for other cameras (eg, front camera and rear camera): for each component of the vision system, if the object in motion is visible in time " and at time center 2, It is expected that for all, σ see ', medium "< ^ < G ' and expect at time " in the part of the correcting image becomes much more than the object in time. Moving away from the visibility (_e(10)Gf仏(8)(4)) in the opposite portion of the image. Thus, the object 34 that becomes visible on the right camera 22 produces light 144237.doc 201117131 for the right front camera to use this one. Due to the difference in viewing angles, the portion of the road 24 that is blocked in the side camera 22 is still visible in the front camera 22, so images from the front camera can be used. Once the chasing car becomes visible in the left side of the right camera and the right part becomes unusable again, a reticle can be created for the front camera so that the other one is not used in this situation (because the obstacle 34 will become more visible) . Since the heading of each camera 22 and the heading of the camera are known and based only on the angle in the mask of the subset camera, a mask is also produced for other cameras. As long as the common part between the frames in the space is being corrected is large enough, it is possible to explicitly create a mask for each camera. However, using only a select subset greatly increases the processing speed and only slightly reduces the results. Therefore, the more the obstacle behavior is consistent with the assumptions stated above, the less the reduction in performance noted. As stated above, the reticle can be interpreted as a weighted image. Black (that is, the grayscale value of 255) means the lowest priority, and white means the highest priority. The first two steps in the mask generation method flow only produce black or white values. As previously suggested, the third step may produce a grayscale value of less than 5, whereby different priorities are given to different cameras based on the mask of the subset camera and the angle of the camera. By this method, it is possible to optimize the production of the positive correction patches 36, 38 obtained from the vertical image in order to improve the visibility of the road surface and the shoulder. Since the same points on the surface 24 of interest can be seen from two different cameras 22 (or from the same camera 22 at different times) at the same time or at different times, improved visibility can be achieved using the concepts of the present invention. Figure 14 illustrates another flow diagram of the technique in which the first slice is provided from the first positive correction 144237.doc • 17-201117131 image (step 42) together with the first tile - step 44), This identification is the first step of the first step (in the discussion of the purpose, any known moving foreground object in the /. The existing part of the horse's swallowing m piece together with its first mask constitutes the outer part of the earth table luxury /, BEM) . In step 46, a new positive = two = system that partially overlaps the first tile is provided to read its location data. Project the second mask to the indication of Pro = two. Similarly, in step %, the second XX H light is placed on the wafer. At step 52, the pitch of the temporary spelling is calculated: 4 this is the Euclidean distance measured from the focus of the camera 22 to the pixel or zone under consideration. The entire coincidence region 32 of one tile and the second tile is compared on a region-by-region basis or possibly pixel by pixel. If the first or destination tile has a free area or pixel, then the corresponding region or pixel from the second, temporary tile is used. This is shown in inquiry 54 and step 56. If the grayscale value of the corresponding pixel or region in the second (ie, temporary) mask is greater than the grayscale value of the corresponding pixel or region in the first mask, the pixel or region from the second or temporary tile is used Replace the pixel or region in the first tile. This is shown in inquiry 58 prior to step 56. If the grayscale values are equal or within a predefined range (as prompted by query 60) then another query is made at 62 to determine if the camera distance of the second, temporary pixel is less than the camera distance of the first, destination pixel. If the second, temporary pixel is acquired from a closer distance, the second, temporary pixel (or region) is copied (i.e., replaced) to the first pixel (or region) according to step 5-6. The mask value (step 64) and camera distance (in step 66) are then updated. An inquiry is made at 68 as to whether the last region or pixel in the consensus zone has been considered. If no, then method steps 52 through 66 are repeated. Once 144237.doc 18-201117131 has been analyzed in this manner from the last pixel (or region) of the consistent region, the updated tile is saved in step 70 along with the updated mask, and the updated tile and the updated The mask becomes part of the mosaic (BEM). Referring to Figures 15 through 18C, the diagram of Figure 14 is graphically represented. In this example, the first tile 38 is represented by the forward pointing camera 22 and the second tile 36 is derived from the angled camera 22. However, it must be understood that the particular orientation of the camera 22 shown in Figure 15 is strictly for illustrative purposes only. The correcting first tile 38 is shown in Figure 16A, while the positive correction tile is shown in Figure 8. The reticle 40 produced for the first tile 38 is shown in Figure 16B and the second tile 36 The reticle 72 is shown in Fig. 17B. In this simplified example, only the moving object 34 (Fig. 7A) is detected in the second tile 36, which corresponds to the reticle μ and the identified moving object 34 m Both of the reticle images are better named in the case. As shown in Fig. 16B, there is no object to be obscured in the picture-like patch % because the moving fir has not been detected in the horizontal image. Therefore, the reticle 40 is completely white. The second splicing mask is not shown in Fig. 17B. Then, the tiles 36, 38 are as shown in i8a: overlapping without a reticle' When the second tile % overlies the first tile", the moving object 34 blurs a portion of the road image that is clearly seen in Figure 16A. Next, the masks 4, 72 are shown as being combined in the figure. By comparing the coincident (ie, overlapping) regions of the first patch 38 and the second tillage sheet 36, the second mask 72 has a grayscale value greater than the grayscale value in the first mask 4〇, Then, the region from the second tile 36 will be used to replace the region of the first tile 38. ^' is the opposite case in this particular example because it corresponds to the two lights 4〇, 72 The comparison of the regions shows the gray scale 144237.doc •19·201117131 r in the second mask 72: the gray scale value of the corresponding region in the f-mask 40. Therefore, using the lower portion of the first open 38 image, As shown in Fig. 18C, a certain part of the animal body 34 of the gram is actually present in (4). The m piece 36 contains the corresponding area region in the first piece 38. The shard 38 is in the first piece 38 There is no corresponding pixel or ^ when 'using the image data from the second tile 36, even if it contains the already ° =: ! 34. The comparison between the zones indicates the first mask and the second mask y = order In the case where the values are substantially equal, the system will evaluate the distance between the individual and the two images. The photographic distance here indicates the positive correction image and the focus of the camera 22. The distance. The minimum photographic distance is considered to be more reliable 'and therefore its image will be over-overset in the overlap η' and the mosaic reticle is updated along with the photo-image distance recorded in the mosaic to enable any subsequent splicing operations The towel, the new positive correction piece will be combined with the recorded reticle data. In this way, the positive correction piece is combined into a mosaic, wherein the overlapping area is selected based on the image content specifically related to the presence of the moving object (4). Techniques, identifying moving objects 34 and then producing a reticle from positively correcting the tiles, can be used to determine which regions of the overlapping tiles should be given priority when producing a mosaic of large surfaces 24 of interest, such as the earth. According to the prior art, the positively corrected patch Τ, . 'α produces a less useful knot because the barrier 34 can cover a portion of the surface 2 of the surface of interest. However, according to the present invention, the use of the reticle Helps select the best available image with horizontal objects (such as lane separation, lane corridors, drain placement, etc.) 144237.doc • 20- 201117131 Therefore, the use of a reticle helps to improve the legibility of the resulting mosaic (BEM), and because such reticle can be produced strictly based on the compared image data, no additional imaging or laser data techniques are required. To identify the moving object 34. The reality is that only a pair of overlapping horizontal (positive corrected) images are needed to produce a bird's eye mosaic (BEM). The movement is detected by a plurality of grayscale differences of the common regions or pixels of the positive correction patch. Object h. Contrary to the change detection in the case of a vertical frame, because the detection is performed in the positive correction space 'so the method can directly distinguish the background from the moving object. Figure 19 shows two alternative applications of the invention' The tiles are generated from images acquired from cameras 122, 222 carried on an aerial vehicle such as a satellite 12 or aircraft 22A. Also, in this case, moving the foreground obstacles 134, 234 can create an obstacle in the resulting image. The direct application of the concepts described herein has the potential to improve the image quality of the resulting mosaics from such aerial images. The foregoing invention has been described in terms of relevant statutory standards, and thus the nature of the description is illustrative and not limiting. The disclosed embodiment can be made obvious to those skilled in the art and fall into the 笳 ... 山# The legal protection given by the present invention [simplified description of the drawing] 辄 can only be wrong by research below Please determine the patent scope. The picture shows a highly simplified description of an action cart that crosses the road and uses a suitable photographic equipment to obtain a series of sequential images. The sequence of the series: GPS positioning data along with adaptive geocoding; 胄 付 付 疋 疋 144 144 144 144 144 144 144 144 144 144 144 144 144 144 201117131 FIGS. 2A-2C illustrate a time sequential view in which an action cart according to the present invention is caught by a foreground moving object (depicted in this case as a sports car); FIG. 3 is a motion chart showing a moving foreground obstacle The time-lapse sequence of the vehicle's moving foreground obstacle partially obscures the image of the desired road surface obtained by a (front-facing) camera but does not make the image of the same surface acquired by a different (backward facing) camera Blur; FIG. 4 is a simplified perspective view of a front facing camera view from the top of a motion graphics vehicle (such as the motion graphics vehicle depicted in FIG. 3), wherein the moving vestibule, P prematurely appears in the front left lane, And the 靡 矣 矣 且 且 且 且 且 且 且 且 且 且 且 且 矣 ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; , wherein the obstacle shown as a blackened portion of the left corner obscures the view of the road surface; w Figure 6 is a patch of the tile depicted in Figure 5 along with the previous first tile (four), configured to show the foreground of the movement The obstacle can be changed from a piece to a piece to change the relative position and can create a view obstacle in one piece, and the piece of sweat does not create a view obstacle in another piece; FIG. 7 depicts Figure 5 is a view of the first tile of Figure 6; Figure 8 is a view of the first tile of Figure 6 in which the movement barrier does not block one portion of the road surface; Figure 9 depicts a reticle of the first region of the tile (Fig. 8); Fig. 1A is a flow chart for describing the mosaic using the method of the present invention; Fig. 11 shows a positive correction for the positive correction tile similar to τ ω of Fig. 5. The film, 1屮 is processed for the purpose of image improvement and $~ 曰 is subdivided into four lines (eight to...; 144237.doc -22- 201117131 Figure 12A is a time chart of the original data collected from the present invention, The column indicates the subdivided area (eight to D) in each tile and the row representation Sequence tiles or images 〇△, ί, ί + &etc.; Figure 12Β is a time diagram of Figure 12, which illustrates the use of behavioral modeling to improve the detection of moving objects in the foreground; Figure 13 is Figure 12 The enlarged view of the area limited to 13; FIG. 14 is a flow chart depicting a sequence of steps for correcting the visibility of the road surface in the positive correction image along the road using the reticle; FIG. 15 is a plurality of cameras mounted A simplified top view of the action cart, two such cameras simultaneously photographing overlapping areas on the surface of interest; Figure 16 depicts a first patch taken by the first camera in front of the direction of the action cart in Figure 15. Figure 16A is a photomask produced for the first tile of Figure 16; Figure 17A is the second phase of the positive correction obtained from the second phase of the action cartographic vehicle in Figure 15; 'Figure 17Β a second reticle generated for the second tile of FIG. 17; FIG. 18A shows the splicing of the first tile and the second tile, wherein the overlapping two tiles are caused by moving the foreground obstacle to make the visible road surface Partially blurred; Figure (10) A comparison between the first reticle and the second reticle, wherein the reticle priority is evaluated and used to determine which portions of the first and second tiles contain more accurate information of the surface of interest; FIG. 18C is as Figure IgA, however, Figure 18C shows the use of improved data obtained by comparison of the first cover to create a mosaic; Figure 19 is a description of the invention by S纟 明 Μ Μ Μ Μ Μ Μ 及Mosaic 144237.doc -23- 201117131 A highly simplified view of the way the application is applied, where the positive correction piece can be derived from satellite imagery and/or aerial imagery. [Main component symbol description] 13 Area 20 Action cart 22 Camera 24 Pavement 26 GPS receiver 28 Satellite 30 Orientation device 32 Overlapping area 34 Moving foreground object 36 Second patch 38 First patch 40 Mask 72 Mask 120 Satellite 122 Camera 134 Moving foreground obstacles 220 Aircraft 222 Camera 234 Moving foreground obstacles A Area B Area 144237.doc -24- 201117131 c Area D Area 144237.doc •25-