不同航測標於影像自動標記之研究

CNU IR > College of Sustainable Environment and Recreation > Dept. of Applied Geoinformatics > Dissertations and Theses > Item 310902800/31862

Please use this identifier to cite or link to this item: https://ir.cnu.edu.tw/handle/310902800/31862

Title:	不同航測標於影像自動標記之研究 Study on Automatic Image Matching Effectiveness of Photogrammetric Target Using Various Shapes
Authors:	陳癸杏
Contributors:	應用空間資訊系林宏麟
Keywords:	影像標記航測標多視立體視覺正射影像 Image marking aerial marker Multi-View Stereo (MVS) Orthomosaic
Date:	2018
Issue Date:	2019-02-27 16:46:30 (UTC+8)
Abstract:	攝影測量航拍之前，必須先佈設航測標，並測量航測標中心點的空間坐標，作為自動化影像標記的基礎，同時也作為影像尺度、基準的約制。傳統數值航空攝影測量的疏點雲自動標記是採用面積基礎之影像標記（Area-Based Image Matching），其方法是利用相鄰影像間局部區塊（視窗）影像灰度值來進行相似性評估（MAD法），或使用統計量（NCC法）或條件約制解算來確認共軛點的位置。面積基礎之標記方法，雖然標記精度高，但對於影像之間如有過大旋轉、過大尺度差異、均調地區或紋理重複區域，很容易造成標記失敗且標記耗時之缺點，因此對航高及姿態角有其嚴格的限制，其航測標的形式也皆為中心正方形標，再視佈標地點配上二至四個翼標之唯一形式。從多視點影像技術出現後，其疏點雲自動標記是採用特徵基礎之影像標記（Feature-Based Image Matching），其方法是利用SIFT演算法進行偵測並提取影像的特徵點進行影像標記並配合RANSAC方法除錯。它可快速獲得大量的特徵點，能自動化的標記出稀疏點雲，標記精度也可達到次像元精度，同時具有旋轉與尺度不變之特性，因此對航高及姿態角不必有何限制，非常適合於UAS航測之特性，也因此發展出多種軟體能自動化標記的航測標形式，如編碼標、圓形標和方形標等。本研究為了解不同形式航測標在影像自動標記的成效，自行設計白底黑圓、黑底白圓、方標、黑底白圓十字標和白底黑元十字標五種非編碼形式的標誌，並採用近景攝影方式進行標記成效分析。經本研究之實驗一確認僅白底黑圓、黑底白圓、方標三種非編碼形式航測標可以自動標記。就短距離拍照的自動標記效果而言，方標較圓形標好，而一般UAS航測的飛行高度皆大於30公尺，因此圓形標仍可適用於UAS航測。經本研究實驗二確認不同拍攝距離所對應不同的GSD情況下，白底黑圓、黑底白圓、方標和編碼標各自標記成功率達90%以上尺寸範圍為：方標16～24個像元、圓標14～18個像元、編碼標37～47個像元之間。若在相同GSD的條件下，航測標所需要的尺寸是圓形標小於方標，方標小於編碼標。然在不考慮GSD的狀況下，拍攝距離在30公尺以下，方標在17個像元以上，二種圓標在15個像元以上，編碼標在37個像元以上，其標記成功率皆為100%。一般UAS航測採用相同的相機和固定的航高拍照，GSD是固定值，故建議採圓標最為適宜。此外，無人機雖易受風速影響，然一般會規畫80%以上的影像重疊率，同一個航測標會在相當數量的影像上存在，若能有90%以上的標記成功率，再配合適當規劃航測標的位置，成果精度應不會受到影響。 Prior to photogrammetry, terrain markers must be set up and the spatial coordinates of the center point for those makers also should be measured as the basis for automatic image markings, and also as a measure of image size and adjustment. The automatic sparse cloud automatic marking of traditional numerical aerial photogrammetry is based on area-based image matching. The method is to use the gray value of the local block (or window) image between adjacent images for relative assessment (MAD method), or use statistical (NCC method) or conditional bundle adjustment to confirm the position of the conjugate point. Although the area-based marking method has higher marking accuracy, however, the failure of marking and the disadvantage of time-consuming marking were caused by the excessive rotation between images, an excessively large scale difference, and uniform areas or texture repeating areas. Therefore, there are strict restrictions on the altitude and attitude angle, and the form of the air survey mark is also the center square mark, and then the unique form of two to four wing tags is attached to the place of the wing mark.After the emergence of multi-view stereo technology, its automatic marking of sparse point clouds is a feature-based image matching. The method is to use SIFT algorithm to detect and extract image feature points for image tagging and RANSAC method debugging. It can quickly obtain a large number of feature points, can automatically mark out the sparse point cloud, the marking accuracy can also reach sub-pixel accuracy, and has the same characteristics of rotation and scale. Therefore, there is no need to limit the altitude and attitude angle, which is very suitable for the characteristics of UAS aerial surveying. As a result, a variety of marker formats such as code marks, round marks, and square marks have been developed for automated software. In this study, for understanding the effectiveness of different types of aerial survey markers in automatic image marking, five types of non-encoded markers are designed: black circles on a white background, white circles on a black background, square symbols, a white cross on a black background, and a black cross on a white background, and use close-range photography for marker effectiveness analysis.In the first experiment of this study, it was confirmed that only three non-encoded forms of white circles, black circles, white circles, and square marks can be automatically marked. In terms of the automatic marking effect of short-distance photographing, square marks are better than circular marks. In general, UAS surveys have a flying height of more than 30 meters, so the circular landmarks can still be applied to UAS aerial surveys. Experiment 2 derive that the success rates of the white circles, black circles, white circles, square marks, and coded marks are 90% or more under different GSD conditions with different shooting distances. The number of pixel is: Square marks range 16 to 24 pixels, the circle mark range 14~18 pixels, the code mark range 37~47 pixels. If the same GSD are used, the required pixel size of the marker is that the circle mark is smaller than the square mark, and the square mark is smaller than the code mark. Meanwhile, shooting distance is below 30 meters without considering GSD, square marker is above 17 pixels, two kinds of circles are above 15 pixels, and coding is above 37 pixels can be reached to 100% effectiveness. It is concluded that the general UAS aerial surveys use the same camera and fixed aerial height photographs, GSD is a fixed value, it is recommended that the circle marker is most suitable. Although drones are disturbed to wind speed, those generally plan to have an image overlap rate more than 80%. The same aerial survey standard will exist on a considerable number of images. If there is more than 90% of the success rate of marking, the aerial drone will be properly planned. The location of the subject, the accuracy of the results should not be affected.
Relation:	電子全文公開日期：2018-06-29，學年度：106， 53頁
Appears in Collections:	[Dept. of Applied Geoinformatics] Dissertations and Theses

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	1410	View/Open

Loading...