Scanning Photogrammetry For Measuring Large Targets In Close Range

Non-metric digital cameras such as the high resolution single-lens reflex cameras are widely used in the fields of photogrammetry and computer vision. However, the field of view (FOV) of digital cameras are limited when long focal lenses are used to measure large targets, since the limitation of electric sensor technology and the cost lead to the format of single frames cannot be large. For high-precision measurements, it is popular to apply oblique photography to increase the intersection angles in close-range photogrammetry. While, large intersection angles make image matching difficult, whereas small ones result in low intersection precision. Therefore, in many measurement methods, multi-view photography is applied to solve this contradiction. Nonetheless, this technique introduces new problems, such as coverage gaps and weak intersection angles, since images are taken manually and only experienced photographers can accomplish the work well. Also, it is difficult to acquire images in order, particularly in large measurement ranges. How to organize images for efficient data processing remains a complicated problem. Therefore, a lot of work to be done to solve these issues.In this paper we employ scanning photography to acquire sequence of images while the camera is rotating at the station, instead of using several cameras to obtain images at one moment or the camera with large FOV such as line-array camera or wide angle lens camera, to enlarge the FOV of one station. The cost of the equipment is low and it is more convenient to conduct measurements. We can change the focus of lens to satisfy different requirements of measurement tasks.1) Scanning photography is a photography approach in which the camera is controlled and rotated around one point to acquire images in multiple rows and columns to enlarge the FOV of one station. The image sequence taken from the same station are referred to as an image matrix. The core idea of scanning photogrammetry is that the view of stations can alternatively be enlarged given that the view of cameras is difficult to enlarge. This method allows stations distributed in order as that in aerial photogrammetry and correspondingly allows standardized data acquired in close range. Therefore, in this paper, we propose the station distribution is set similar to the flight strip in aerial photogrammetry to facilitate data acquisition and processing. Also, a photo scanner consisting of a non-metric camera, a rotation platform, and a controller is developed for convenient data acquisition without coverage gap. Once the operator sets the required parameters and specifies the ground coverage of the station, this machine can automatically control the camera to rotate in both horizontal and vertical directions and obtain images as designed. It can reduce error rate due to lack of experience of operators.2) The traits of data obtained by scanning photography are that, overlap between the ground coverages of adjacent image matrices is known and larger than 60% based on station distribution, whereas the overlap between stereo original images from different stations is unknown and may be less than 50%. Therefore, a modified triangulation is proposed, in which the images synthesized from image matrices are processed as the bridge. We generate the free network of synthetic images first, and the overlap between the original images can be identified indirectly base on this free network.3) When establishing the free network of original image, we can avoid matching all images with one another, which will greatly improve the efficiency of data processing. The reason of synthetic images are not used for bundle adjustment is that information loss is unavoidable during image synthesis. When data is acquired by photo scanner, the rotation angles of each image in both horizontal and vertical direction will be recorded by the scanner. And the offset values of the center of rotation and the center of camera are small compared with the photo distance when large engineering measurements are measured using this photo scanner. Thus, we can ignore the offset and calculate the location and posture of each original images in the free network of synthetic images based on the rotation angles recorded. These exterior elements of original images are inaccurate but are valid as the initial data for bundle adjustment convergence. And, this method is more robust than methods that take relative orientation among original images to obtain the exterior elements of original images, because that the overlap between original stereo images from adjacent stations may be less than 50% and insufficient for relative orientation.4) Given the differences of translation values of the original images acquired from the same station are very small in scanning photogrammetry, the traditional bundle adjustment model in which exterior elements of each image are considered as independent parameters and will not be suitable for processing the data. To overcome this problem, we propose a new bundle adjustment model which considers the exterior elements of each station, the rotation angles of each original image in both horizontal and vertical direction, and the location and posture of the basis image with rotation angles as zeros in the station coordinate system as the independent parameters. And the exterior elements of each original image can be calculated according to these parameters. We confirmed that this method is conducive to the convergence of bundle adjustment for scanning photography data.5) The feasibility and precision of the proposed method are validated by tests on two fields using 200mm and 300mm lenses. The results confirm that even with a small amount of control points, the developed scanning photogrammetry can steadily achieve the measurement accuracy better than 5mm at distances ranging from 40 m to 230 m.