Automatic High Precision Registration Between Aerial Images And LiDAR Data Without Ground Control Points

Digital photogrammetry, especially digital aerial photogrammetry, has developed a whole theory system after several decades of development. As one of geographical information acquiring methods, it has a greate advantage in texture show. In recent years, low altitude photogrammetry and unconventional aerial photogrammetry have played a very important role in emergency response against the natural disasters such as earthquake, etc. Meanwhile, airborne laser system, which usually called LiDAR (Light Detetion And Ranging), has achieved great development too. It offers a new method to acquire high-resolution geographic information. Compared with aerial image, LiDAR data is rich in spectrum and textural information, the greatest advantage is that it can give three-dimensional information of plenty ground points directly. In a word, both kinds of data have its advantages and disadvantages, and they are complementary to each other. With the development of relative technology, the integration of digital aerial photogrammetry and LiDAR is an inevitable trend. However, in order to arrive it, one of key issues is to realize automatic high precision registeration between aerial imageas and LiDAR data. That means the point on aerial image is corresponding to LiDAR data, and LiDAR point is corresponding to aerial image.In this thesis, using the experience of aerotriangulation operation, an automatic high precision registration method of aerial images and LiDAR data is proposed in condition of large area for aerial images and LiDAR data without ground control points (GCPs). The kernel part is to extract and match dense feature points on aerial images, get the three dimensional coordinates corresponding to the feature points with intersection. So, the problem of registration between aerial images and LiDAR data is turned to regiser image matching points to LiDAR points. The Iterative Closest Point Algorithm (ICP) is one of effective methods between laser points sets. However, the ICP needs good initial value, and time consuming of searching the corresponding points is very much. After solving these two problems, ICP is used to image matching points and LiDAR points. The iterative idea is adopted during the registration of aerial images and LiDAR data. After registering dense points cloud on control images during iterative procedure, good registered points are choosed as control points for bundle block adjustment. Finally, the orientation elements of images are carried out, and the registration of aerial images and LiDAR data is realized.For unconventional aerial photogrammetry which degree of overlap can not meet the requirement of the block aerotriangulation, stero image dense matching is adopted, and making use of ICP to register matching points to LiDAR points, then orientation elements of images are carried out by resection with filtered registration points, and so registration between aerial images and LiDAR data in condition of unconvertional aerial photogrammetry comes true.Some problems involved in the registering process are as follows:(1) the initial orientation elements of imagesTake the Position and Orientation System (POS) data as the initial exterior orientation elements of the images if there are, or after tie points of aerotriangulation matching automatically based on Scale Invariant Feature Transform (SIFT), getting a few of rough control points from LiDAR data manully, and then making adjustment to calculate the initial orientation elements of images.(2)multi-images dense matching under constraint of LiDAR and initial orientation elementsAccording to the plan of setting GCPs during aerial photogrammetry, some images are taken as control ones. With meshing control images and extracting dense Harris feature points, multi-images matching under constraint of LiDAR and initial orientation elements is put into practice. At the same time, the points in the water areas are deleted making use of LiDAR water areas character, finally high precision image matching points are acquired.(3) Registration of image matching points and LiDAR points.The iterative algorithm is applied to register image matching points and LiDAR points. In each iterative operation, firstly,3D coordinates of image matching points are calculated by intersectin with the last orientation parameters of images after adjustment (in the first iterative operation, making use of initial orientation elements or POS data). Secondly, the key corresponding points of the two point clouds are choosed by the nearest distance, and image matching points are registered to LiDAR points according to Iterative Closest Point Algorithm. After taking high precision registration points as control points for bundle block adjustment, the new orientation elements of images are carried out. Go on with the iterative operation unless the precision satisfy demand. At last, the orientation elements of images in the LiDAR coordinate reference frame are acquired, and automatic high precision registration of aerial images and LiDAR data without GCPs is realized.(4) Calculating orientation elements of images by spatial resection with registration pointsIn situation of unconventional aerial photogrammetry, after registration between matching points and LiDAR points, high precision registration points are sifted for spatial resection to calculate orientation elements of images, and the precision is evaluated.The registration experiments of unconventional aerial images and LiDAR data is indicated that this registration method which by registering image matching points and LiDAR points has a greate advantage in higter automation and precision compare with manual registration.For conventional aerial images, orientation elements of images in the LiDAR coordinate reference frame are carried out according to aerotriangulation operation procedure, and automatic high precision registration of aerial images and LiDAR data is implemented in this way.