1375182 (1) 九、發明說明 【發明所屬之技術領域】 本發明係有關廣角影像之影像接圖；更明確地，係有 關一種可快速旦動態地找岀多重影像之適當接圖點的方法 【先前技術】 藉由將個別影像無縫地接合起來以產生廣角（全景） 影像是一種常見的習知技術。此技術通常僅侷限於遠距物 體之影像的接圖，或者只能針對某固定距離的目標物來執 行接圖，而無法針對具有不同距離的動態物體之複雜影像 (或短距物體之影像）找出適當接圖點來執行接圖。因爲 此種複雜影像之接圖需要有拍攝物體與相機間之距離的相 關資訊。欲求得上述距離之正確數値，需要相當繁複且耗 時的計算工作。但是，對於即時處理動態影片而言，爲了 提供即時且平順的輸出，將無法接受相當繁複且耗時的計 算工作。因此，希望能以更快速且簡單的方法來執行接圖 操作。 【發明內容】 本發明之目的係提供一種能夠動態且快速地找出最佳 接圖點以將個別影像無縫地接圖成一廣角（全景）影像的 方法。利用此快速動態接圖點搜尋方法，得以減少習知技 術之複雜計算而並未犧牲接圖點搜尋之準確性。 -4- (2) 1375182 爲達成上述目的，依據本發明之一種快速動 搜尋方法包含：根據相機角度以決定一第一影像 圖區及一第二影像之可能接圖區：計算出針對不 距離之座標轉換矩陣，以便將該第一影像中之座 該第二影像；選擇具有一特定尺寸之區塊；選擇 區塊之搜尋視窗，其中該搜尋視窗不超出該可能 執行該第一影像與該第二影像間之區塊比對程序 數候選接圖點；及於該複數候選接圖點中決定一 點。 依據本發明之特徵，該區塊比對程序包含： 影像之該可能接圖區內隨機地選擇複數預定接圖 pn);利用該座標轉換矩陣以將該複數預定接圖 pn)轉換至該第二影像之該可能接圖區內而取得 接圖點（ΡΊ至ρ’π):計算該複數預定接圖點（ )爲中心之區塊與包括該複數相應接圖點（Ρ’ι 3 複數搜尋視窗（W1至Wn )內的所有區塊間的第 差異絕對値之和）値；及決定每一該複數搜尋i 至Wn)內具有最小第一 SAD値之複數候選接圖 A„ )。 依據本發明之特徵’該決定一最佳接圖點之 :以該複數候選接圖點（Ai至An)爲接圖點執 影像與該第二影像之接圖，以取得複數接圖影像 ):於該複數接圖影像（1>至In)內選取一接圖 計算每一該複數候選接圖點（A,至An)針對該 態接圖點 之可能接 同的指定 標轉換至 包括複數 接圖區； 以找出複 最佳接圖 於該第一 點（p I至 點（P!至 複數相應 :Pi 至 Pn i p，n)之 —SAD ( 視窗（W, 點（A ,至 步驟包含 行該第一 (I!至 U 判斷區； 接圖判斷 -5- (3) 1375182 區之個別第二SAD (差異絕對値之和）値；及選取該複數 候選接圖點（A,至An )中具有最小第二SAD値者爲一最 佳接圖點。 依據本發明之特徵，該第一及第二SAD値之計算係 根據像素値。 依據本發明之特徵，進一步包含在決定可能接圖區之 後執行彩色至灰階轉換及邊緣檢測程序。 【實施方式】 圖1係一種多眼視頻相機系統之槪略示意圖。如圖1 中所示，一多眼視頻相機系統100包含N個鏡頭（川至 LN)，其個別地將拍攝影像傳送至N個影像信號處理單元 (ISPUi至ISPUN) 110。該等影像信號處理單元110係處 理該等鏡頭所拍攝之影像以增進影像之品質。經個別影像 信號處理單元（ISPU,至ISPUN) 110處理後之影像被傳 • 送至一廣角影像產生器120。廣角影像產生器120包含一 多重影像自動曝光（AE)比對單元122，用以減少不同光 線環境下所拍攝之影像間的色差問題；及一接圖單元124 ' ，用以針對處理後之影像進行接圖點搜尋及接圖操作。最 後可輸出一廣角影像》 接下來，將參考圖2至圖5以說明依據本發明之快速 動態接圖點搜尋方法，其中圖2係顯示依據本發明之快速 動態接圖點捜尋方法200的流程；圖3係說明多眼視頻相 機系統中以不同角度鏡頭所拍攝的兩個影像；圖4係說明 -6- (4) (4)1375182 依據本發明之快速動態接圖點捜尋方法的區塊比對（ block matching)程序；及圖5係說明如何從複數候選接 圖點中選出一最佳接圖點。 首先，如圖3所示，假設欲執行接圖操作之雨影像爲 一中心影像C及一側邊影像S，其中之陰影部分分別爲中 心影像C及側邊影像S之重疊區域（於下文中，稱爲可能 接圖點區PRc及PRs )，這些可能接圖點區係根據不同的 相機角度以及物體距離而調整。由於物件距離難以獲得， 所以通常僅藉由相機角度以取得可能接圖點區。於這些可 能接圖點區內，將根據相機角度、物體距離以及所欲的接 圖點精確度以決定捜尋視窗，如圖3中所示之W，以便於 這些搜尋視窗W內找出最佳接圖點（將詳述於下文中） 。應注意的是，相較於習知技術針對所有可能接圖點執行 區塊比對，依據本發明之接圖點搜尋方法僅於搜尋視窗內 執行區塊比對，因而得以減少計算複雜度並加速接圖點之 搜尋。 參考圖2，於步驟ST 2 10，執行影像切割（分別針對 中心影像C與側邊影像S)，以去除影像中之多餘部分而 只留下可能接圖點區PRc及PRs。於步驟ST 220，執行彩 色至灰階轉換及邊緣檢測程序。此步驟之目的在於減少因 拍攝時之不同角度、光線等因素所造成之色差的問題，以 增進接圖點搜尋之精確度。然而，應注意的是，若已知並 無色差問題存在於欲執行接圖操作的兩影像之間，則無須 執行步驟ST 22 0 »當然，亦可逕行省略此步驟以便達成更 (5) (5)1375182 快速的接圖點搜尋。 接下來，於步驟ST 230,事先計算出針對不同的指定 距離之座標轉換矩陣，以便後續於不同影像（中心影像C 與側邊影像S )間之座標轉換。於步驟ST 240，隨機地選 取可能接圖點區PRc中之數個預定接圖點P (如圖3及圖 4所示）。於步驟ST 2 5 0，根據所欲的接圖點精確度以選 擇區塊尺寸及捜尋視窗尺寸。應注意的是，在一定的捜尋 範圍內，選定之區塊尺寸越小，則需做的區塊比對量越大 ，故找到的最佳接圖點更爲精確。此外，搜尋視窗的尺寸 越大，則可能漏掉最佳接圖點之機率越小。 然後，於步驟ST 260，藉由計算像素値之SAD (差異 絕對値之和）來執行區塊比對程序以找出候選接圖點。以 下將參考圖4以進一步說明步驟ST 260中所執行之區塊 比對程序。 首先’應瞭解的是，影像之最小單位爲像素（pixel) ，因此’影像中之每個點均代表一個像素，且均具有一特 定的像素値。如圖4中所示，假設可能接圖點區PRc中之 各個點的座標爲（X，y)，且已於上述步驟ST 240隨機地選 取了 η個預定接圖點P (或？，至pn)。接著，利用上述 步驟ST 230中所求得之座標轉換矩陣以計算出可能接圖 點區PRS中之各對應點P，（或P’,至P’n )的座標（x，，y’） 。舉例而言’假設針對距離d,之上述座標轉換矩陣爲一 2 x2之矩陣，則P’之座標（x’，y’）爲 (6) (6)1375182 (x，，y，）= (x，y) χ Μι 接著，分別以P及p’爲中心’根據所欲的接圖點精確 度以選擇區塊尺寸，亦即’以p及p ’爲中心點分別取得一 BxB之區塊，其中B稱之爲區塊尺寸。舉例而言’於圖4 之範例中，係選擇一 3x3(B = 3)之包含9個點（像素） 的區塊，其中各像素均具有一特定的像素値（如Pi之區 塊所示）。再次強調，在一定的搜尋範圍內，區塊尺寸越 小（B之値越小），則需做的區塊比對量越大，雖較爲耗 時，但能找到較爲精確的最佳接圖點。 此時即根據下列方程式以計算P爲中心之區塊與P’爲 中心之區塊間的SAD (差異絕對値之和）値： L5/2J L££2j SAD(x',y')= Y Y \fs {x'+i, y'+j) -fc(x + i, y + y)| (1) i=-|_B/2j>=-Ls/2j 其中i及j爲整數； fc(x，y)爲可能接圖點區PRC中之座標（X，y)的像素値 :及 fs(x’，y’）爲可能接圖點區PRS中之座標（X，，y’）的像素 値。 此SAD(x’，y’）値即用來表示可能接圖點區PRs中之 點P’與可能接圖點區PRC中之點p的近似程度，SAD(x，， y ’）之値越小則表示兩者越近似。 至此’上述方程式（1)可算出可能接圖點區PRC中之 -9- (7) (7)1375182 點P與可能接圖點區PRS中之其對應點P’的近似程度，例 如，圖4中所示之P,與P, ’間的近似程度' P 2與P 2 ’間 的近似程度、或Pn與Pn’間的近似程度。然而’由於最佳 接圖點不一定剛好位於P’點上，故爲了更精確地搜尋接圖 點，需以Pi’至Pn’爲中心選取一大小適中的捜尋視窗W ( 或\^!至Wn)，進一步於此搜尋視窗W中找出最接近預 定接圖點P之最佳接圖點。再次強調’選定之捜尋視窗的 尺寸越大，則可能漏掉最佳接圖點之機率越小。舉例而s ，假設圖4所示之捜尋視窗w,爲包括卩’！之25個點（像 素），則需於此Wi內針對所有可能的區塊（共9個3x3 之區塊），再次利用上述方程式（1)，計算出可能接圖點區 PRC中之Pi與這些可能的區塊（除了已計算過之P’1爲 中心的區塊以外）之間的近似程度。此時再將所有的 SAD(x，，y，）値（於本範例中爲9個）比較後，取出其中最 小的値，即可判斷該値所代表之點爲最接近p 1之點（於 下文中稱之爲候選接圖點A)，於圖4中以At表示。 之後，重複上述步驟以計算出W2中與p2最接近的候 選接圖點A2、...W^中與Pm最接近的候選接圖點An·/ 及Wn中與Pn最接近的候選接圖點An。 最後，於步驟ST 270，藉由計算所有候選接圖點A 之SAD (差異絕對値之和）來決定最佳接圖點。以下將參 考圖5以說明步驟ST 270如何在候選接圖點八！至 中決定出何者爲最佳接圖點。 如圖5所示，假設以預定接圖點Pi及候選接圖點Al -10- (8) 1375182 分別爲中心影像C之及側邊影像S之接圖點來執行接圖， 則獲得如該圖下半部所示之一接圖影像h。於此接圖影像 Ii中，以該接圖點Ai (於接圖後與P,爲同一點）爲中心 選取一塊寬度爲D而高度爲Η之接圖判斷區J，如圖5中 所示之陰影部分。根據下列方程式以計算候選接圖點Α| 針對該接圖判斷區J之SAD (差異絕對値之和）値：1375182 (1) IX. Description of the Invention [Technical Field of the Invention] The present invention relates to an image connection of a wide-angle image; more specifically, to a method for quickly finding an appropriate connection point of a plurality of images. Prior Art It is a common conventional technique to seamlessly join individual images to produce wide-angle (panoramic) images. This technique is usually limited to the connection of images of distant objects, or can only be performed for a fixed-distance target, but not for complex images of dynamic objects with different distances (or images of short-range objects). Find the appropriate connection point to perform the connection. Because the connection of such complex images requires information about the distance between the subject and the camera. To find the correct number of distances above, considerable computational and time consuming calculations are required. However, for instant processing of dynamic movies, in order to provide immediate and smooth output, it will not be able to accept quite complicated and time-consuming calculations. Therefore, it is desirable to perform the connection operation in a faster and simpler manner. SUMMARY OF THE INVENTION It is an object of the present invention to provide a method for dynamically and quickly finding an optimal connection point to seamlessly map individual images into a wide-angle (panoramic) image. With this fast dynamic map point search method, complex calculations of conventional techniques can be reduced without sacrificing the accuracy of the map point search. -4- (2) 1375182 In order to achieve the above object, a fast motion search method according to the present invention includes: determining a first image area and a possible image area of a second image according to a camera angle: calculating a distance a coordinate conversion matrix for locating the second image in the first image; selecting a block having a specific size; selecting a search window of the block, wherein the search window does not exceed the possible execution of the first image and the The block between the second images compares the number of program candidate points; and determines a point in the plurality of candidate points. According to a feature of the present invention, the block comparison program includes: randomly selecting a plurality of predetermined connection patterns pn in the possible connection area of the image; using the coordinate conversion matrix to convert the plurality of predetermined connection patterns pn) to the first The possible image is connected to the image area to obtain the connection point (ΡΊ to ρ'π): the block corresponding to the complex predetermined connection point ( ) is calculated and the corresponding connection point including the complex number (Ρ'ι 3 plural Searching for the sum of the absolute differences between all the blocks in the window (W1 to Wn)); and determining the complex candidate map A„) having the smallest first SAD値 in each of the complex searches i to Wn). According to the feature of the present invention, the optimal connection point is determined by using the complex candidate connection point (Ai to An) as the connection point image and the connection image of the second image to obtain a plurality of connection images) : selecting a map in the complex image (1> to In) to calculate a possible combination of each of the plurality of candidate points (A, to An) for the state map point to include a complex number Connect the picture area; to find the complex optimal picture at the first point (p I to point (P! to complex phase) :Pi to Pn ip,n)-SAD (Windows (W, point (A, to step contains the line first (I! to U judgment area; figure judgement -5- (3) 1375182 area of the second SAD (the sum of the absolute differences); and selecting the smallest candidate node (A, to An) having the smallest second SAD is an optimal connection point. According to the features of the present invention, the first The second SAD is calculated according to the pixel. According to a feature of the present invention, the color to gray scale conversion and edge detection process is further performed after determining the possible connection area. [Embodiment] FIG. 1 is a multi-eye video camera system. BRIEF DESCRIPTION OF THE DRAWINGS As shown in FIG. 1, a multi-eye video camera system 100 includes N lenses (Chuan to LN) that individually transmit captured images to N image signal processing units (ISPUi to ISPUN) 110. The image signal processing unit 110 processes the images captured by the lenses to enhance the quality of the image. The images processed by the individual image signal processing unit (ISPU, to ISPUN) 110 are transmitted to a wide-angle image generator 120. The wide-angle image generator 120 includes a The multiple image automatic exposure (AE) comparison unit 122 is configured to reduce the color difference between the images captured in different light environments; and a connection unit 124' for searching and connecting the processed images. Figure operation. Finally, a wide-angle image can be output. Next, a fast dynamic map point search method according to the present invention will be described with reference to FIG. 2 to FIG. 5, wherein FIG. 2 shows a fast dynamic map point search according to the present invention. The flow of the method 200; FIG. 3 is a diagram illustrating two images taken by different angle lenses in the multi-eye video camera system; FIG. 4 is a diagram showing the fast dynamic connection point of the -6-(4) (4) 1375182 according to the present invention. The block matching procedure of the seek method; and FIG. 5 illustrates how to select an optimal map point from the plurality of candidate map points. First, as shown in FIG. 3, it is assumed that the rain image to be subjected to the connection operation is a central image C and a side image S, wherein the shaded portions are overlapping regions of the central image C and the side image S (hereinafter , referred to as possible connection point areas PRc and PRs), these possible connection points are adjusted according to different camera angles and object distances. Since the object distance is difficult to obtain, it is usually only by the camera angle to obtain the possible connection point area. In these possible connection points, the search window will be determined according to the camera angle, the object distance and the desired point accuracy, as shown in Figure 3, so that the search window W can find the most Good map points (more on this below). It should be noted that the block search method according to the present invention performs block alignment only in the search window, thereby reducing computational complexity and performing block alignment for all possible connection points. Accelerate the search of the map points. Referring to Fig. 2, in step ST 2 10, image cutting (for the center image C and the side image S, respectively) is performed to remove redundant portions of the image leaving only the possible dot regions PRc and PRs. In step ST220, a color to grayscale conversion and edge detection procedure is performed. The purpose of this step is to reduce the chromatic aberration caused by different angles, light and other factors during shooting to improve the accuracy of the connection point search. However, it should be noted that if it is known that no chromatic aberration problem exists between the two images for which the drawing operation is to be performed, then step ST 22 0 is not required to be performed. Of course, this step can also be omitted to achieve more (5) ( 5) 1373182 Fast connection point search. Next, in step ST230, coordinate conversion matrices for different specified distances are calculated in advance for subsequent coordinate conversion between different images (center image C and side image S). In step ST240, a plurality of predetermined connection points P (as shown in Figs. 3 and 4) in the possible connection dot area PRc are randomly selected. In step ST 2 50, the block size and the window size are selected according to the desired pixel accuracy. It should be noted that within a certain range of search, the smaller the selected block size, the larger the block comparison needs to be made, so the best connection point found is more accurate. In addition, the larger the size of the search window, the less likely it is that the best connection point will be missed. Then, in step ST260, the block alignment program is executed by calculating the SAD of the pixel ( (the sum of the absolute differences 差异) to find the candidate map point. The block comparison procedure executed in step ST 260 will be further explained below with reference to FIG. First, it should be understood that the smallest unit of the image is a pixel, so each point in the image represents one pixel and each has a specific pixel 値. As shown in FIG. 4, it is assumed that the coordinates of the respective points in the possible pixel area PRc are (X, y), and n predetermined connection points P (or ?) have been randomly selected in the above-mentioned step ST240. Pn). Next, using the coordinate transformation matrix obtained in the above step ST 230 to calculate the coordinates (x, y') of the corresponding points P, (or P', to P'n) in the possible connection point region PRS. . For example, 'assuming that the above coordinate transformation matrix is a matrix of 2 x 2 for the distance d, then the coordinate (x', y') of P' is (6) (6) 1375182 (x,, y,) = ( x, y) χ Μι Next, respectively, with P and p' as the center, respectively, according to the accuracy of the desired map point, select the block size, that is, 'p and p' as the center point to obtain a BxB block respectively. Where B is called the block size. For example, in the example of FIG. 4, a block of 9 dots (pixels) of 3×3 (B=3) is selected, wherein each pixel has a specific pixel 値 (as shown by the block of Pi). ). Once again, in a certain search range, the smaller the block size (the smaller the B is), the larger the block size to be made, although it is time consuming, but it can find the more accurate and best. Connect the map. At this time, according to the following equation, the SAD (the sum of absolute differences) between the block centered on P and the block centered on P' is calculated: L5/2J L££2j SAD(x', y')= YY \fs {x'+i, y'+j) -fc(x + i, y + y)| (1) i=-|_B/2j>=-Ls/2j where i and j are integers; fc (x, y) is the pixel 値 of the coordinates (X, y) in the possible connection point region PRC: and fs(x', y') is the coordinate (X, y') in the possible connection point region PRS The pixel is 値. This SAD(x',y')値 is used to indicate the approximate degree of the point P' in the possible connection point region PRs and the point p in the possible connection point region PRC, after SAD(x,, y ') The smaller the value, the closer the two are. So far, the above equation (1) can calculate the approximate degree of the -9-(7) (7) 1375182 point P in the possible connection point area PRC and its corresponding point P' in the possible connection point area PRS, for example, The degree of approximation between P, P, 'approximation degree' P 2 and P 2 ', or the degree of approximation between Pn and Pn'. However, since the best connection point is not necessarily at the P' point, in order to search for the connection point more accurately, it is necessary to select a moderately sized window W (or \^!) centered on Pi' to Pn'. To Wn), further find the best connection point closest to the predetermined connection point P in the search window W. Again, the larger the size of the selected window, the less likely it is that the best point will be missed. For example, let's assume that the window w is shown in Figure 4, including 卩'! For 25 points (pixels), it is necessary to use the above equation (1) for all possible blocks (a total of 9 3x3 blocks) in this Wi, and calculate the Pi in the possible connection point region PRC. The degree of approximation between these possible blocks (except for the blocks where P'1 is centered). At this point, compare all the SAD(x, y,) 値 (9 in this example), and take out the smallest 値, and then judge the point represented by the 为 to be the closest to p 1 ( This is referred to hereinafter as the candidate map point A) and is indicated by At in FIG. After that, the above steps are repeated to calculate the candidate connection pattern closest to Pn among the candidate connection points An·/ and Wn closest to Pm in the candidate connection points A2, ...W^ closest to p2 in W2. Point An. Finally, in step ST 270, the optimum map point is determined by calculating the SAD (the sum of the absolute differences 差异) of all candidate map points A. Reference will be made to Figure 5 below to illustrate how step ST 270 is at the candidate map point eight! It is up to the middle to decide which is the best point to connect. As shown in FIG. 5, it is assumed that the connection is performed by using the predetermined connection point Pi and the candidate connection points A10 (8) 1375182 as the connection points of the central image C and the side image S, respectively. One of the images shown in the lower half of the figure is connected to the image h. In the connection image Ii, a connection judgment area J having a width D and a height Η is selected centering on the connection point Ai (the same point after P and the same point), as shown in FIG. The shaded part. Calculate the candidate map point Α according to the following equation | Determine the SAD of the zone J for this joint (the sum of the absolute differences):
i=0 j=0 (2) 其中i及j爲整數； fs(x’，y’）爲側邊影像S中之座標（x’，y’）的像素値；及 fc(x, y)爲中心影像c之座標（X，y)的像素値。 如此所計算出之SAD値係表示候選接圖點A於接線 L (如圖5中所示）兩邊之對稱位置上之像素値的差異絕 對値之和。該SAD値越小則代表接線L兩邊對稱位置上 的像素値越接近，亦即，接圖的效果越好。 接著利用上述方程式（2)，重複地針對A2 ( P2) 、A3 (P3 ) '…、Αη·, ( Ρ „-,)及 Αη ( Ρη )計算出其他相 應接圖影像h至Ιη之個別的SAD値，再從總共η個SAD 値（分別針對A 1至A n )中取出最小的，即可決定該値所 代表之候選接圖點Α爲最佳接圖點。 在決定了最佳接圖點之後，即根據此最佳接圖點來執 行接圖操作。接著，執行alpha混和（alpha blending)處 理程序以使影像間之接圖處的顏色變化較爲平順，藉此減 -11 - (9) 1375182 少接圖處之不自然感。此時即可產生一無縫的廣角（全景 )影像。 發明之效果 欲達成無縫的接圖影像需要得知從相機至拍攝物體間 之距離（即，影像深度）。然而，欲藉由影像之計算以得 知影像深度是極困難的工作。此外，習知用於視頻相機影 g 像之接圖點搜尋方法在應用於高解析度的多重影像時將耗 費龐大的運算量及時間。 本發明係事先計算出針對不同的指定距離之座標轉換 - 矩陣。如此得以減少接圖點搜尋區域，進而加速接圖點之 搜尋。此外，本發明利用改良的區塊比對方式以增進接圖 點搜尋之精確度。因此，本發明不僅減少了習知接圖點搜 尋方法之龐大的運算量及時間，亦同時克服了要求得知影 像深度之問題，而可動態且快速地找出最佳接圖點。 【圖式簡單說明】 圖1係多眼視頻相機系統之槪略示意圖。 圖2係依據本發明之快速動態接圖點搜尋方法的流程 圖。 . 圖3係一示意圖，其顯示多眼視頻相機系統中以不同 角度之鏡頭所拍攝的兩個影像。 圖4係一圖形，其說明依據本發明之快速動態接圖點 搜尋方法的區塊比對（block matching)程序。 -12- (10) (10)1375182 圖5係一圖形，其說明如何從複數候選接圖點中選出 一最佳接圖點。 【主要元件符號說明】 100 :多眼視頻相機系統 110:影像信號處理單元 120:廣角影像產生器 122 :多重影像自動曝光（AE)比對單元 124 :接圖單元i=0 j=0 (2) where i and j are integers; fs(x',y') is the pixel 座 of the coordinates (x',y') in the side image S; and fc(x, y) The pixel of the coordinate (X, y) of the center image c. The SAD system thus calculated represents the sum of the absolute differences of the pixel pupils at the symmetric positions on both sides of the candidate interface point A at the wiring L (as shown in Fig. 5). The smaller the SAD 则 is, the closer the pixel 对称 on the symmetrical position on both sides of the wiring L is, that is, the better the effect of the connection. Then, using the above equation (2), repeatedly calculate the respective corresponding image h to Ιη for A2 (P2), A3 (P3) '..., Αη·, ( Ρ „-,) and Αη ( Ρη ) SAD値, and then take the smallest of the total of η SAD 値 (for A 1 to A n respectively), then determine the candidate map point represented by the Α as the best connection point. After the point, the drawing operation is performed according to the optimal drawing point. Then, the alpha blending processing program is executed to make the color change at the connection between the images smoother, thereby reducing -11 - (9) 1375182 Less unnaturalness in the picture. A seamless wide-angle (panoramic) image can be produced at this time. The effect of the invention To achieve a seamless image, you need to know the distance from the camera to the subject. (ie, image depth). However, it is extremely difficult to know the depth of the image by the calculation of the image. In addition, the conventional method for searching the image of the video camera image is applied to high resolution. Multiple images will consume a lot of computation and time. Firstly calculate the coordinate conversion-matrix for different specified distances. This reduces the search point search area and accelerates the search of the connection points. In addition, the present invention utilizes the improved block comparison method to enhance the connection point search. Accuracy. Therefore, the present invention not only reduces the huge amount of computation and time of the conventional image point searching method, but also overcomes the problem of requiring the image depth to be known, and can find the optimal connection point dynamically and quickly. BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a schematic diagram of a multi-eye video camera system. Fig. 2 is a flow chart of a fast dynamic picture point searching method according to the present invention. Fig. 3 is a schematic diagram showing a multi-eye video. Two images taken at different angles of the lens in the camera system. Figure 4 is a diagram illustrating a block matching procedure for a fast dynamic map search method in accordance with the present invention. -12- (10 (10) 1375182 Figure 5 is a graphic showing how to select an optimal connection point from a plurality of candidate connection points. [Main component symbol description] 100: Multi-eye video camera system 110: Shadow Image signal processing unit 120: Wide-angle image generator 122: Multiple image automatic exposure (AE) comparison unit 124: Connection unit
-13--13-