China 
Africa 
Explore chapters and articles related to this topic
Visual Perception
Published in Konar Amit, Artificial Intelligence and Soft Computing, 2018
Finding the corresponding points in images is generally referred to as the correspondence problem [3]. There are many approaches to handle the correspondence problem. In this book we will first determine the 3-D points of an object from its 2-D image points and then for more than two images determine the correspondence by measuring the shortest Euclidean distance between the point sets of the two or more images. If the 2-D to 3-D mapping of the points are satisfactory and the images consist of common points, then determining the correspondence between the points of two or more images is not a complex problem. For determining the 3-D points from their 2-D images we, in this book, will use Kalman filtering [1]. However, before introducing Kalman filtering, we briefly outline the minimal representation of 2-D lines, 3-D lines and 3-D planes. With a minimal representation, we can extract 3-D points from multiple 2-D points in different images by using Kalman filtering. We use Kalman filtering because it has an advantage of recursively operating on incoming data stream like 2-D points from n number of images. The more is the value of n, the better will be the accuracy of the results. The other least square estimators, unlike Kalman filtering, demand all the data set together; so the user has no choice to control the level of accuracy at the cost of computational time. In Kalman filtering, one can observe the improvement in accuracy in the estimation of the parameter of lines or planes and accordingly decide about the submission of new data points.
Stereoacuity and optics
Published in Pablo Artal, Handbook of Visual Optics, 2017
To elucidate how the visual system can calculate the disparity map of a visual scene is an extraordinarily complex problem in which the final mechanisms are still unclear. The problem of calculating the disparity map of a visual scene is also known as the stereo correspondence problem (Howard and Rogers 1995, 2002; Marr 1980; Marr and Poggio 1976), since to know the disparity map is equivalent to determining, given a point on the retina, to what point on the other retina it corresponds in order to produce a simple, unified vision. In a normal scene, from the points of the image on a retina, there may be millions of pairs corresponding to those of the other retina but only one real valid correspondence that generates the disparity map associated with the visual image being viewed. The problem of stereo correspondence is clearly manifested with random-dot stereograms (RDS) (explained immediately in the following). In the solution of the stereo correspondence problem, the visual information at different spatial frequencies constitutes an essential point in the solution of this problem (Howard and Rogers 1995, 2002; Marr 1980).
3D imaging
Published in Michael O’Byrne, Bidisha Ghosh, Franck Schoefs, Vikram Pakrashi, Image-Based Damage Assessment for Underwater Inspections, 2019
Bidisha Ghosh, Michael O’Byrne, Franck Schoefs, Vikram Pakrashi
The next key step that is thoroughly addressed is how to solve the stereo correspondence problem, that is, the problem of finding the same points in the left and right camera images. This problem is non-trivial when the imagery is collected in challenging visibility conditions characterized by non-uniform lighting and limited range visibility. A robust stereo matching algorithm based on belief propagation (BP) is described, which is well-suited for poor visibility conditions. With knowledge of the intrinsic and extrinsic camera parameters of the stereo system, and having found corresponding points in the left and right camera images, this chapter proceeds to show how a 3D point cloud can be created via triangulation.
A novel depth estimation approach based on bidirectional matching for stereo vision systems
Published in Advanced Robotics, 2020
The traditional stereo vision technique typically consists of four steps: camera calibration, rectification, correspondence problem and triangulation. However, the mainstream of stereo vision research in the past two decades has been focused on correspondence problem which is an evergreen research question in computer vision despite the innumerable amount of contributed algorithms [2–4]. Stereo correspondence problem popularly known as stereo matching is an under-constrained task, i.e. there is no sufficient information available to guarantee finding correct and unique matches in all real-world stereo images. The sources of this ill-posed problem include textureless regions, repetitive patterns, occlusion, reflection and slanted surfaces to cite a few. In the past decade, quite a number of promising methods [5–24] have been proposed for stereo matching. Many researchers rely on a collection of some useful constraints of which epipolar constraint is the most popular to reduce the complexity of the problem. Epipolar constraint depends on the geometry of the stereo cameras and the stereo images acquisition process. This constraint facilitates stereo matching by reducing potential 2D search space to 1D space. A typical stereo matching technique undergoes four steps: matching cost computation [4], cost aggregation [5–11], disparity optimization and disparity refinement [13–18].
Photogrammetric refinement of LiDAR-derived building roof contours
Published in International Journal of Image and Data Fusion, 2018
The proposed method is based on two main steps. First, straight lines representing the sides of a selected building roof contour are extracted from the image and projected onto corresponding LiDAR-derived building polyhedron faces by using a line-based photogrammetric model. It is worthy noting that, for each selected building, the straight lines extracted from the image must be matched to corresponding polyhedron roof faces. This correspondence problem can be solved algorithmically or visually. In this paper neither the straight-line extraction nor the matching between straight lines and planar roof faces is accomplished by specific algorithms. Section 2.1 presents the line-based photogrammetric model. Second, the refined contour of the polyhedron is determined by connecting every pair of adjacent projected lines, resulting in a new set of contour vertices. Section 2.2 describes the method for determining the refined contour of the polyhedron.
An underwater lighting and turbidity image repository for analysing the performance of image-based non-destructive techniques
Published in Structure and Infrastructure Engineering, 2018
Michael O’Byrne, Franck Schoefs, Vikram Pakrashi, Bidisha Ghosh
The correspondence problem is often difficult because of ambiguous correspondences that can lead to false matches. This is especially problematic for uniform surface types whereby the lack of distinct features causes a high number of vague matches. Such scenarios are particularly likely to arise in underwater settings due to the poor visibility conditions. Additionally, another problem arises when correspondence cannot be reliably established because a region is occluded in one of the images. The correspondence problem has been extensively researched and a wide range of stereo matching algorithms have been proposed. (Scharstein & Szeliski, 2002) provide a taxonomy of the most notable algorithms.