Unordered Image Sets Research Articles

DeepImSeq: Deep image sequencing for unsynchronized cameras

Consider a set of n images of a dynamic scene captured using multiple hand-held devices. The order in which these images are captured is unknown. For n images, there can be n! possible arrangements, which makes this problem extremely challenging. In this work, we address the problem of sequencing such a set of unordered images in its temporal order. We propose an LSTM-based deep neural network which addresses this problem in an end-to-end manner. The network takes the set of images as input and outputs their order of capture. We formulate the problem as a sequence-to-sequence mapping task, in which each image is mapped to its position in the ordered sequence. We do not provide any other information to the network apart from the input images. We show that the proposed approach obtains the state-of-the-art results on the standard dataset. Further, we show through experimental results that the network learns better when the target sequence is reversed.

Structure from motion for ordered and unordered image sets based on random k-d forests and global pose estimation

In this paper, we present a new fast and robust method for structure from motion (SfM) for data sets potentially comprising thousands of ordered or unordered images. Our work focuses on the two most time-consuming procedures: (a) image matching and (b) pose estimation. For image matching, a new method employing a random k-d forest is proposed to quickly obtain pairs of overlapping images from an unordered set. After that, image matching and the estimation of relative orientation parameters are performed only for pairs found to be very likely to overlap. For pose estimation, we use a two-stage global approach, separating the determination of rotation matrices and translation parameters; the latter are computed simultaneously using a new method. In order to cope with outliers in the relative orientations, which global approaches are particularly sensitive to, we present a new constraint based on triplet loop closure errors of rotation and translation. Finally, a robust bundle adjustment is carried out to refine the image orientation parameters.We demonstrate the potential and limitations of our pipeline using various real-world datasets including ordered image data acquired from UAV (unmanned aerial vehicle) and other platforms as well as unordered data from the internet. The experiments show that our work performs better than comparable state-of-the-art SfM systems in terms of run time, while we achieve a similar accuracy and robustness.

Efficient tree-structured SfM by RANSAC generalized Procrustes analysis

This paper proposes a tree-structured structure-from-motion (SfM) method that recovers 3D scene structures and estimates camera poses from unordered image sets. Starting from atomic structures spanning the scene, we build well-connected structure groups, and propose RANSAC generalized Procrustes analysis (RGPA) to glue structures in the same group. The grouping-aligning operations hierarchically proceed until the full scene is reconstructed. Our work is the first attempt of using GPA for modern 3D reconstruction tasks. RGPA is able to merge multiple structures at a time and automatically identify outliers. The reconstruction tree is much more compact and balanced than previous hierarchical SfM methods and has a very shallow depth. These advantages, along with the resulting removal of intermediate bundle adjustments, lead to significantly improved computational efficiency over state-of-the-art SfM methods. The cameras and 3D scene can be robustly recovered in the presence of moderate noise. We verify the efficacy of our method on a variety of datasets, and demonstrate that our method is able to produce metric reconstructions efficiently and robustly.

THE PERFORMANCE EVALUATION OF MULTI-IMAGE 3D RECONSTRUCTION SOFTWARE WITH DIFFERENT SENSORS

Abstract. Today, multi-image 3D reconstruction is an active research field and generating three dimensional model of the objects is one the most discussed issues in Photogrammetry and Computer Vision that can be accomplished using range-based or image-based methods. Very accurate and dense point clouds generated by range-based methods such as structured light systems and laser scanners has introduced them as reliable tools in the industry. Image-based 3D digitization methodologies offer the option of reconstructing an object by a set of unordered images that depict it from different viewpoints. As their hardware requirements are narrowed down to a digital camera and a computer system, they compose an attractive 3D digitization approach, consequently, although range-based methods are generally very accurate, image-based methods are low-cost and can be easily used by non-professional users. One of the factors affecting the accuracy of the obtained model in image-based methods is the software and algorithm used to generate three dimensional model. These algorithms are provided in the form of commercial software, open source and web-based services. Another important factor in the accuracy of the obtained model is the type of sensor used. Due to availability of mobile sensors to the public, popularity of professional sensors and the advent of stereo sensors, a comparison of these three sensors plays an effective role in evaluating and finding the optimized method to generate three-dimensional models. Lots of research has been accomplished to identify a suitable software and algorithm to achieve an accurate and complete model, however little attention is paid to the type of sensors used and its effects on the quality of the final model. The purpose of this paper is deliberation and the introduction of an appropriate combination of a sensor and software to provide a complete model with the highest accuracy. To do this, different software, used in previous studies, were compared and the most popular ones in each category were selected (Arc 3D, Visual SfM, Sure, Agisoft). Also four small objects with distinct geometric properties and especial complexities were chosen and their accurate models as reliable true data was created using ATOS Compact Scan 2M 3D scanner. Images were taken using Fujifilm Real 3D stereo camera, Apple iPhone 5 and Nikon D3200 professional camera and three dimensional models of the objects were obtained using each of the software. Finally, a comprehensive comparison between the detailed reviews of the results on the data set showed that the best combination of software and sensors for generating three-dimensional models is directly related to the object shape as well as the expected accuracy of the final model. Generally better quantitative and qualitative results were obtained by using the Nikon D3200 professional camera, while Fujifilm Real 3D stereo camera and Apple iPhone 5 were the second and third respectively in this comparison. On the other hand, three software of Visual SfM, Sure and Agisoft had a hard competition to achieve the most accurate and complete model of the objects and the best software was different according to the geometric properties of the object.

An efficient method for organizing unordered images in 3D reconstruction

Modeling 3D objects from unordered image data sets is being researched for several years. In this paper we investigate the problem of inefficiency in organizing unordered image sets into clusters of related. With the increase in number of images, computation for pairwise matching becomes more expensive. Practically considerable algorithms are more focused on the organization of images. Instead of saving time from the process of arrangement of views, we emphasize on the time-consuming part “comparison”. This process is ahead of establishment of the relationship among views which includes building of the spanning tree or skeletal graph. The proposed algorithm achieves similar results to that of pairwise matching algorithm. More importantly it computationally saves more time by the introduction of a requirement before pairwise comparison which reduces computation time for two-view matching. The method we present has been tested on image sets both on the Internet and shot by ourselves. And it has been observed that more widely separated views lead to much better results.

A Fast and Robust Framework for Semiautomatic and Automatic Registration of Photographs to 3D Geometry

We present a simple, fast, and robust complete framework for 2D/3D registration capable to align in a semiautomatic or completely automatic manner a large set of unordered images to a massive point cloud. Our method converts the hard to solve image-to-geometry registration task in a Structure-from-Motion (SfM) plus a 3D/3D alignment problem. We exploit a SfM framework that, starting just from an unordered image collection, computes an estimate of the camera parameters and a sparse 3D geometry deriving from matched image features. We then coarsely register this model to the given 3D geometry by estimating a global scale and absolute orientation using two solutions: a minimal user intervention or a stochastic global point set registration approach. A specialized sparse bundle adjustment (SBA) step, that exploits the correspondence between the sparse geometry and the fine input 3D model, is then used to refine intrinsic and extrinsic parameters of each camera. Output data is suitable for photo blending frameworks to produce seamless colored models. The effectiveness of the method is demonstrated on a series of synthetic and real-world 2D/3D Cultural Heritage datasets.

DETECTION OF CRITICAL CAMERA CONFIGURATIONS FOR STRUCTURE FROM MOTION

Abstract. This paper deals with the detection of critical, i.e., poor or degenerate camera configurations, with a poor or undefined intersection geometry between views. This is the basis for a calibrated Structure from Motion (SfM) approach employing image triplets for complex, unordered image sets, e.g., obtained by combining terrestrial images and images from small Unmanned Aerial Systems (UAS). Poor intersection geometry results from a small ratio between the baseline length and the depth of the scene. If there is no baseline between views, the intersection geometry becomes undefined. Our approach can detect image pairs without or with a very weak baseline (motion degeneracy). For the detection we have developed various metrics and evaluated them by means of extensive experiments with about 1500 image pairs. The metrics are based on properties of the reconstructed 3D points, such as the roundness of the error ellipsoid. The detection of weak baselines is formulated as a classification problem using the metrics as features. Machine learning techniques are applied to improve the classification. By taking into account the critical camera configurations during the iterative composition of the image set, a complete, metric 3D reconstruction of the whole scene could be achieved also in this case. We sketch our approach for the orientation of unordered image sets and finally demonstrate that the approach is able to produce very accurate and reliable orientations.

Statistical Computations on Grassmann and Stiefel Manifolds for Image and Video-Based Recognition

In this paper, we examine image and video-based recognition applications where the underlying models have a special structure—the linear subspace structure. We discuss how commonly used parametric models for videos and image sets can be described using the unified framework of Grassmann and Stiefel manifolds. We first show that the parameters of linear dynamic models are finite-dimensional linear subspaces of appropriate dimensions. Unordered image sets as samples from a finite-dimensional linear subspace naturally fall under this framework. We show that an inference over subspaces can be naturally cast as an inference problem on the Grassmann manifold. To perform recognition using subspace-based models, we need tools from the Riemannian geometry of the Grassmann manifold. This involves a study of the geometric properties of the space, appropriate definitions of Riemannian metrics, and definition of geodesics. Further, we derive statistical modeling of inter and intraclass variations that respect the geometry of the space. We apply techniques such as intrinsic and extrinsic statistics to enable maximum-likelihood classification. We also provide algorithms for unsupervised clustering derived from the geometry of the manifold. Finally, we demonstrate the improved performance of these methods in a wide variety of vision applications such as activity recognition, video-based face recognition, object recognition from image sets, and activity-based video clustering.

Horizon matching for localizing unordered panoramic images

There is currently an abundance of vision algorithms which, provided with a sequence of images that have been acquired from sufficiently close successive 3D locations, are capable of determining the relative positions of the viewpoints from which the images have been captured. However, very few of these algorithms can cope with unordered image sets. This paper presents an efficient method for recovering the position and orientation parameters corresponding to the viewpoints of a set of panoramic images for which no a priori order information is available, along with certain structure information regarding the imaged environment. The proposed approach assumes that all images have been acquired from a constant height above a planar ground and operates sequentially, employing the Levenshtein distance to deduce the spatial proximity of image viewpoints and thus determine the order in which images should be processed. The Levenshtein distance also provides matches between imaged points, from which their corresponding environment points can be reconstructed. Image matching with the aid of the Levenshtein distance forms the crux of an iterative process that alternates between image localization from multiple reconstructed points and point reconstruction from multiple image projections, until all views have been localized. Periodic refinement of the reconstruction with the aid of bundle adjustment, distributes the reconstruction error among images. The approach is demonstrated on several unordered sets of panoramic images obtained in indoor environments.

3D curve structure reconstruction from a sparse set of unordered images

A flexible and portable approach to reconstructing 3D curve structures of free-form objects from a sparse set of unordered images taken by a hand-held CCD camera is presented. To make the whole pipeline robust and accurate, we mark the interested curves on the object to make them visually outstand from the background. Some specifically designed markers are distributed around the object as well. Followed by an automatic camera pose recovery and camera lens distortion correction step, a user-friendly semi-automatic method is put forward to extract the curves in the images. The curve correspondences in stereo image pairs are then optimally established and the 3D curve structures are recovered by using the stereo triangulation method. Experiments show that the method is applicable for reverse modeling of freeform objects with moderate accuracy requirements.