China
Africa
Explore chapters and articles related to this topic
Real-time 3D reconstruction using SLAM for building construction
Published in Konstantinos Papadikis, Chee S. Chin, Isaac Galobardes, Guobin Gong, Fangyu Guo, Sustainable Buildings and Structures: Building a Sustainable Tomorrow, 2019
The main functionality of visual odometry is to estimate camera motion based on images taken. Images contain a matrix of brightness and color. It is difficult to estimate camera motion directly from a matrix of data. Therefore, some representative points are extracted from the image first and these points will remain the same after changing the positon and orientation of the camera. These representative points are called Features in visual SLAM. The camera position and orientation will be estimated by using calculating the changes of these Features. The corners and edges in the image are taken as features since they are more distinguishable among different images. However, when the camera position is far away from one object, the corner may not be recognized. Therefore, in this research, Oriented FAST and Rotated BRIEF (ORB) is used to detect Features based on four characteristics: repeatability, distinctiveness, efficiency and locality (Rublee et al. 2011). The ORB feature is composed of two parts: the key point and descriptor. Its key point is called Oriented Features from Accelerated Segment Test (FAST), which is the FAST corner point (Rosten et al. 2006). Its descriptor is called the Binary Robust Independent Elementary Features (BRIEF) (Calonder et al. 2010). After the ORB features are calculated, a fast library for approximate nearest neighbors (FLANN) is used to match ORB features in each image (Muja et al. 2014). Finally, the camera positon and orientation is calculated using PnP (Perspective-n-Point) with the positon data of the ORB features (Lepetit et al. 2013).
Three-dimensional CityGML building models in mobile augmented reality: a smartphone-based pose tracking system
Published in International Journal of Digital Earth, 2021
Christoph Blut, Jörg Blankenbach
To maintain the flexibility of device-based pose tracking, cameras have found an increasing application for realizing optical pose tracking. For instance, White and Feiner (2009) include 6 degree of freedom optical marker tracking in SiteLens, for displaying virtual 3D building models. The fiducial markers are distributed across the environment and are captured by the mobile camera to provide known reference points for correcting the device pose. But like stationary external tracking systems, marker tracking also requires preparing the environment in advance, limiting the versatility of the AR systems. Therefore, efforts have been made to integrate existing physical objects, such as buildings, referred to as natural features, in optical pose tracking. Vacchetti, Lepetit, and Fua (2004), Wuest, Vial, and Stricker (2005), Reitmayr and Drummond (2006), Lima et al. (2010), Choi and Christensen (2012) and Petit, Marchand, and Kanani (2013) show virtual 3D model-based solutions using edge-matching methods. The defining edges of the 3D models are utilized to search for corresponding 2D edges of the physical objects in camera images and matched with these to derive a pose with the Perspective-n-Point (PnP) algorithm. Reitmayr and Drummond (2006) use their system to overlay virtual wireframe models over physical buildings. The drawbacks of these approaches are that the objects must be captured continuously by the camera to estimate poses, restricting mobility, and that, typically, prepared wireframe models are required.
A novel algorithm for pose estimation based on generalized orthogonal iteration with uncertainty-weighted measuring error of feature points
Published in Journal of Modern Optics, 2018
Ju Huo, Guiyang Zhang, Jiashan Cui, Ming Yang
Pose estimation plays an important role in the field of photogrammetry, computer graphics and robotics. Within the last decade, the methodologies for pose estimation have attracted considerable attention in visual measurement system (1,2). A basic problem of pose estimation is to estimate the rotation and translation of the moving object relative to the target coordinate system. It obtains the relative motion information using the coordinates of moving object in the world coordinate system and its imaging coordinates projected by camera, which can be attributed to solve the problem of perspective-n-point (PnP) motion parameters (3,4). The pose estimation of stereo vision usually makes use of large number of points (n ≥ 3) to complete the moving target matching, and the imaging model of target projection is established to solve the relative motion parameters. As for the solution to perspective-n-point problem, the investigations are not rare and often solved using an iterative optimization algorithm. It is commonly practiced to construct a minimized objective function and the optimal solution obtained by minimizing the error function via loop iterations, see (5) for instance.