臺灣博碩士論文加值系統

近年來隨著數位影像處理技術與數位相機的快速發展，攝影測量的自
動化系統正逐步地實現，其中近景攝影測量學者最希望達成的目標就是人
工標點的自動偵測、定位與辨識。目前對制式標點（如：圓形標）之偵測
與定位技術已趨成熟並廣為應用，然而傳統的方法卻只認標點不認標號，
尚須以人工方式給予標號，又由於標點型式均相同，因此在找尋共軛標點
時容易混淆。本研究將兩個圓形標及兩位數字結合成一個複合式標點，除
了運用影像處理技術對標點進行偵測、定位外，並利用辨識標號來分辨各
個標點。 本文所建議的作法分為下面四個步驟：首先，
利用圓形標偵測元來偵測所有可能的圓形標，並利用最小二乘匹確定位出
圓形標中心點的精確位置。其次，量取圓心連線長度來決定比例尺，進而
擷取出標號的位置。第三，則是將擷取出的數字轉正，並重新取樣至數字
樣本視窗的大小。最後計算數字視窗與數字樣本視窗的標準化互相關係數
，以決定該標點的標號。

根據上述方法吾人建立了一套認標的自動化程式，並以數位相機實際攝影
進行三項實驗。第一項實驗拍攝單純白色背景下的人工標點，在此情形下
所有標點都可以正確無誤地辨識出來；第二項實驗拍攝階梯上的人工標點
，其中階梯上方的標點由於比例尺較小，造成無法辨識或錯誤的辨識結果
；第三項實驗是由輕微傾斜的側面拍攝圓形標，而實驗結果顯示，當傾斜
角度小於30度時，圓形標偵測元仍能有不錯的偵測率。

The invention of digital camera stimulates photogrammetrists
to pursue anautomatic system of photogrammetry. In
close-range photogrammetry,automatically detecting,
identifying and positioning targets are primary goalsto be
reached in recent years. Currently proposed methods utilize
targets ofregular shape, such as a round-shape target, which can
be detected and locatedeasily in a digital image. However,
a target like this kind can not beidentified among a number of
others in an image. It is therefore difficult tofind the
conjugate targets in some overlapped images. In this paper, two
roundshape targets and two digits are composed to be a
distinguishable target fromothers.By using the proposed image-
processing techniques, a compound target isnot only detectable
and locatable but also distinguishable by interpreting thetwo-
digit subimage to obtain its identification number.
There are 4 steps in the proposed method. First, a
circular-targetdetector is used to search all possible
targets, and locates the centersprecisely by using the least-
squares matching technique. Second, the scale ofthe target is
calculated by measuring the length between the centers of thetwo
round-shape targets, so that the boundaries of the two
digits can bedefined. Third, subimages of the digits are
extracted and resampled to be thesame orientation and
resolution as the provided templates of standard digits.Finally,
the subimages of the digits are compared with the templates of
thestandard digits to solve the identification number.

Three experiments were conducted to show the feasibility of
the proposedwhite color background. All the targets were
successfully and correctlyidentified in this experiment. The
second experiment is designed to findtargets in different
scales, so that several sets of targets attached in thevertical
sides of a stairs were imaged and tested. Targets tend
to bemisinterpreted when their scales are getting small. The
third experiment isdesigned to identify targets in an oblique
image. This experiment shows thatcorrect results can be
obtained when the image angle is smaller than 30degrees.