China
Africa
Explore chapters and articles related to this topic
Spatial Registration
Published in Praveen Kumar, Jay Alameda, Peter Bajcsy, Mike Folk, Momcilo Markus, Hydroinformatics: Data Integrative Approaches in Computation, Analysis, and Modeling, 2005
A reference coordinate system is usually defined by a user and depends on a particular application. For example, it would be preferable to choose a reference coordinate system for a sequence of aerial images (see Figure 17.2) that conforms to a map projection of other datasets. This type of registration is also called image rectification. Imagery that was transformed by the orthogonal projection (perpendicular to the ground plane) and was freed from geometric distortions is called orthoimagery. In many other applications, such as, video registration or 3D medical volume reconstruction, any image in a set of processed images can be chosen as the reference coordinate system.
Dynamic and static object detection and tracking in an autonomous surface vehicle
Published in Ships and Offshore Structures, 2020
Elham Omrani, Hossein Mousazadeh, Mahmoud Omid, Mehdi Tale Masouleh, Hamid Jafarbiglu, Yousef Salmani-Zakaria, Ashkan Makhsoos, Farshid Monhaseri, Ali Kiapei
Image rectification uses Epipolar constraint to project images onto a common image plane without any distortion in order to simplify finding matching points between images. Stereo matching is the process of finding the similarity between left and right images by the sum of squared difference (SSD) method to calculate disparity. The best correspondence for each pixel can be determined by minimum SSD.
Turbulent free-surface monitoring with an RGB-D sensor: the hydraulic jump case
Published in Journal of Hydraulic Research, 2021
Daniel B. Bung, Brian M. Crookston, Daniel Valero
As previously noted, the technology employed in this camera is active stereoscopy (Barnard & Fischler, 1982; Kanade & Okutomi, 1991). The camera includes the Intel® RealSense Vision Processor D4 (specifications available at https://www.mouser.com/pdfdocs/Intel_Vision_Processor_D4_ProductBrief.pdf) with a proprietary depth algorithm (that receives input from the optical module comprised of the two IR cameras, the infrared projector, and the RBG camera for colour). Fundamentally, the device uses two cameras with a fixed separation distance (known as the baseline) to capture at a precise moment a left and right image pair, which is compared for correspondence. This process involves two images taken along epipolar lines, image rectification where left and right images are reprojected on a common virtual plane, a proprietary filtering algorithm to remove failed matching due to occlusion, and computing disparity throughout the FOV. This means that any shifts required to properly match the right image to the left image are computed (number of pixels shifted = disparity). Through triangulation, a depth per pixel is derived from: where Xres= horizontal image resolution, and HFOV = the horizontal FOV (∼90°). In order to compute depth resolution, Eq. (1) can be differentiated with respect to disparity. Intel (Grunnet-Jepsen et al., 2019b) notes that depth resolution is a function of the smallest disparity step or subpixel. Because this algorithm is proprietary, specific details are not available; however, documentation (https://dev.intelrealsense.com/docs) provided by the manufacturer describes various aspects of performance, set-up, sensitivities, and camera settings users may adjust and select in the user interface (RealSense Viewer and LibRealSense). LibRealSense also includes a selection of preset parameter combinations or setting profiles that are also available in the user interface (see RGB-D Camera Settings).