China
Africa
Explore chapters and articles related to this topic
Unmanned Aerial System Applications to Coastal Environments
Published in David R. Green, Billy J. Gregory, Alex R. Karachok, Unmanned Aerial Remote Sensing, 2020
Francesco Mancini, Marco Dubbini
In SfM (SFM) methodology, the interior orientation parameters are unknown or derived as approximate values, from the exchangeable image file header of JPEG images collected by a UAV. Approximate values are required in the SfM (SFM) approach, to facilitate the search for conjugate points. A successive refinement of the interior orientation parameters is determined later in the bundle block adjustment, which incorporates the option for camera self-calibration. In the computer vision strategies for image analysis, the search for conjugate points detected in the source photographs is based on point detectors and descriptors of the scale-invariant feature transform (SIFT) type (Lowe 2004). In SfM (SFM) processes, a least squares adjustment is implemented to detect incorrect matches, which are discarded. The whole process produces the so-called 3D sparse point clouds. A successive step, where a multi-view stereo algorithm is performed, produces 3D dense point clouds, and optionally, meshed products and orthophotographs. Moreover, in dense stereo-matching methods, the recent implementation of semi-global matching methods (Hirschmüller 2008) has provided new advances in the generation of 3D point clouds and DEMs. In Figure 6.1, a simplified overview of UAV-based mapping processes with SfM (SFM) methodologies is provided.
Road Scene 3D Reconstruction
Published in Jian Chen, Bingxi Jia, Kaixiang Zhang, Multi-View Geometry Based Visual Perception and Control of Robotic Systems, 2018
Jian Chen, Bingxi Jia, Kaixiang Zhang
For comparison, the widely used semi-global matching method [93] is employed to obtain disparity maps. The resulting disparity map is shown in Figure 5.8. Note that the disparity map is only semi-dense, i.e., there exist many holes that are mismatched and eliminated by stereo validation. This phenomenon gets worse for textureless roads, as will be shown in the fifth scene in Figure 5.16. In real world, the road regions are likely to be textureless or repetitively textured, arising problems for classical stereo matching algorithms, as they are designed for general scenes and usually need to search over a large region to determine the correspondences and there exist many ambiguities. In comparison, the 3D reconstruction method, in this chapter, is designed specifically for road scenes. The regions of interest are believed to be continuous and near the reference plane; thus, the search region can be greatly reduced to deal with repetitive textures. Textureless regions can also be well dealt with by the spline-based approximation of the parallax information. Dense and smooth reconstruction result is generated from the two-view images based on which the road region is easy to be detected without the need of complex detection strategies.
Depth sensing with disparity space images and three-dimensional recursive search
Published in Automatika, 2018
Miroslav Rožić, Tomislav Pribanić
While many research endeavors have tried to improve upon the accuracy of DP-based methods, these approaches have generally yielded an increase in computational complexity, which has been in turn addressed by high-performance hardware architectures [12]. Semi-global matching [5] (SGM), widely used in many real-time and real-world applications, makes extensive use of DP to compute the disparity for each pixel in the image by estimating the disparity for multiple paths intersecting for every pixel of the image. For each image pair a tensor of all possible disparities is computed, and for each path at each pixel a DSI is formed as a cross-section of the tensor in the path direction (horizontal, vertical, or diagonal). For methods such as SGM and many methods derived from it, optimizing the DP and DSI computation would yield significant improvements in execution time and memory footprint.
Colour-weighted rank transform and improved dynamic programming for fast and accurate stereo matching
Published in The Imaging Science Journal, 2023
Mohamed Hallek, Randa Khemiri, Ali Aseere, Abdellatif Mtibaa, Mohamed Atri
Figure 7 indicates that the graph cut algorithm gives acceptable results in the Middlebury benchmark, but this method is very costly and its implementation in real time is practically impossible. Traditional dynamic programming and semi-global matching are two efficient solutions for optimization and disparity calculation. They can outperform the graph cut approach in terms of accuracy for some images as shown in Figure 8. In addition, the proposed optimization approach denoted that improved DP produces smooth disparity maps and shows least errors for all images. It is the most accurate optimization technique.