State of the art in high density image matching

Abstract

Image matching has a history of more than 50 years, with the first experiments performed with analogue procedures for cartographic and mapping purposes. The recent integration of computer vision algorithms and photogrammetric methods is leading to interesting procedures which have increasingly automated the entire image-based 3D modelling process. Image matching is one of the key steps in 3D modelling and mapping. This paper presents a critical review and analysis of four dense image-matching algorithms, available as open-source and commercial software, for the generation of dense point clouds. The eight datasets employed include scenes recorded from terrestrial and aerial blocks, acquired with convergent and normal (parallel axes) images, and with different scales. Geometric analyses are reported in which the point clouds produced with each of the different algorithms are compared with one another and also to groundtruth data.It is quite evident that even with our past progress, we have only scratched the surface of the possibilities in the use of photogrammetry. ) at different scales. Complex scenes and objects can be surveyed and reconstructed using a large set of images with very satisfactory results (Fig. 1). In particular, methods for dense point-cloud generation (dense image matching) are increasingly available for professional and amateur applications such as 3D modelling and mapping, robotics, medical imaging, surveillance, tracking and navigation.Due to the availability of a number of different low-cost and open-source software systems, automated 3D reconstruction methods are becoming very popular. Nevertheless, the metrological and reliability aspects of the resulting 3D measurements and modelling should not be ignored, particularly if the community wishes to adopt such solutions not only for quick 3D modelling and visualisation but also for accurate measurement purposes. To this end, clear accuracy statements, benchmarking and evaluations must be carried out.This paper presents a critical review and analysis of selected dense image-matching algorithms. The algorithms considered are from both the commercial and open-source domains. The datasets adopted for the testing (Table I and Fig. 3) include terrestrial and aerial image blocks, acquired with convergent and normal (parallel axes) images at different scales and resolution. With respect to other reported benchmarking datasets, the imagery considered here is of higher resolution and it covers more complex scenes. Moreover, the evaluations presented are performed on the raw output of the matching (that is, on the point cloud) and not at the mesh level. The algorithms are evaluated according to their ability to produce dense and high-quality 3D point clouds, as well as according to computation time. Geometric analyses are reported, in which the point clouds produced with each of the different algorithms are compared with one another and also to ground-truth data. Laser Scanning or Photogrammetry?Since 2000, range sensors, both airborne and terrestrial, ...

Topographic Maps of Mount Etna’s Summit Craters, updated to December 2015

“…This process allowed generating a first 3D point cloud that includes all the portions of Etna. Using this approach, the surface measurement and reconstruction is formulated as an energy function minimization problem, using pyramidal processing (Remondino, Spera, Nocerino, Menna, & Nex, 2014). This strategy, which could be defined as hierarchical, is followed in order to optimize the results in terms of speed and quality; first, the best homologous points are found using a highly subsampled set of images that allows producing rough layout data that can be refined step-by-step on images with gradually improved resolution (pyramidal approach) and moreover enables a reduction of the research area for each pixel.…”

Landslide Fissure Inference Assessment by ANFIS and Logistic Regression Using UAS-Based Photogrammetry

“…The null deviance was calculated as 630.20, by comparing the null model with the theoretical model (also known as the saturated model), the model deviance was calculated as 340.57 by comparing the full model, including all predictors with the theoretical model, while chi-square was determined as 289.63 using Equation (11). The set of predictors significantly improved the model fit that can, therefore, be concluded from the regression results, as the model deviance is significantly smaller than the null deviance.…”

“…Lightweight, low-cost, high-resolution compact digital cameras of about 18-24 megapixels can be mounted easily to the drones producing high-quality aerial images from convenient flight heights. Then, the acquired images are evaluated to obtain high-density 3D data with low-cost photogrammetric software, e.g., Photomodeler Scanner [6], Pix4D Mapper [7], Agisoft PhotoScan [8], or with open-source photogrammetric software, e.g., surface reconstruction from imagery (SURE) [9], or patch-based multi-view stereo (PMVS) [10] photogrammetric software [11].…”

Structure from motion (SfM) processing of UAV images and combination with terrestrial laser scanning, applied for a 3D-documentation in a hazardous situation

“…Performance of this technique depends on the image quality and collection mode, as well as lighting conditions and the presence of sharp edges, which could cause some problems in feature detection due to a poor signal-to-noise ratio (Remondino et al 2014;Dandois et al 2015;Tannant 2015;Bevilacqua et al 2016).…”

“…Laser scanning and photogrammetry are surely very effective surveying methodologies as for application flexibility, due to their ability to yield high-density geometric and radiometric data (Garnero et al 2010;Rinaudo et al 2012;Caroti, Camiciottoli et al 2013;Guarnieri et al 2013;Remondino et al 2014;Caroti et al 2015).…”