Fig 4 - uploaded by Simone Gasparini
Content may be subject to copyright.
3 Line images for the equiangular camera model. The black curve is the image of any line in the focal plane i.e., the plane containing the optical center and perpendicular to the optical axis. It is naturally a circle and corresponds to the hemispheric part of the field of view. The other curves show the images of lines whose interpretation planes (plane spanned by a line and the optical center) form angles of 70, 50, 30, and 10 degrees with the optical axis, respectively. The limiting case of 0 degrees corresponds to lines that are coplanar with the optical axis and whose images are lines going through the distortion center. Although in general, line images are not algebraic curves, the parts within the hemispheric field of view, i.e., within the black circle, can be relatively well approximated by conics.

3 Line images for the equiangular camera model. The black curve is the image of any line in the focal plane i.e., the plane containing the optical center and perpendicular to the optical axis. It is naturally a circle and corresponds to the hemispheric part of the field of view. The other curves show the images of lines whose interpretation planes (plane spanned by a line and the optical center) form angles of 70, 50, 30, and 10 degrees with the optical axis, respectively. The limiting case of 0 degrees corresponds to lines that are coplanar with the optical axis and whose images are lines going through the distortion center. Although in general, line images are not algebraic curves, the parts within the hemispheric field of view, i.e., within the black circle, can be relatively well approximated by conics.

Source publication
Article
Full-text available
This survey is mainly motivated by the increased availability and use of panoramic image acquisition devices, in computer vision and various of its applications. Different technologies and different computational models thereof exist and algorithms and theoretical studies for geometric computer vision ("structure-from-motion") are often re-develope...

Similar publications

Article
Full-text available
The number of researchers utilising imagery for the 3D reconstruction of underwater natural (e.g., reefs) and man-made structures (e.g., shipwrecks) is increasing. Often, the same procedures and software solutions are used for processing the images as in-air without considering additional aberrations that can be caused by the change of the medium f...
Article
Full-text available
In this article, we present an interactive method to enable the 3D modelling of buildings from a set of uncalibrated images. Matching of homologous points and constraints (geometric ones in our examples) are used to proceed to the 3D model estimation. First, we introduce a bundle adjustment algorithm that hides the pose parameters by using classica...
Article
Full-text available
We present a method for fully automatic 3D reconstruction from a pair of weakly calibrated images in order to deal with the modeling of complex rigid scenes. A 2D triangular mesh model of the scene is calculated using a two-step algorithm mixing sparse matching and dense motion estimation approaches. The 2D mesh is iteratively refined to fit any ar...
Conference Paper
Full-text available
In this paper, we present our recent sensor fusion approaches to obtain high-quality 3D information. We first discuss two fusion methods that combine geometric and photometric information. The first method, multiview photometric stereo, reconstructs the full 3D shape of a target object. The geometric and photometric information is efficiently fused...

Citations

... First, a pinhole camera is used, a common non-distorted model employed in game rendering. Second, we analyze a low-fidelity, single-parameter model based on the field of view (FOV) model [28], which can capture some distortion but makes simplistic assumptions about the geometry of the lens. Lastly, we consider a radial model commonly used in computer vision which serves as the highest-fidelity model considered; with well calibrated parameters, this was found to give distortions accurate to within 1% of the true distortion of the actual camera. ...
Preprint
Full-text available
Modeling cameras for the simulation of autonomous robotics is critical for generating synthetic images with appropriate realism to effectively evaluate a perception algorithm in simulation. In many cases though, simulated images are produced by traditional rendering techniques that exclude or superficially handle processing steps and aspects encountered in the actual camera pipeline. The purpose of this contribution is to quantify the degree to which the exclusion from the camera model of various image generation steps or aspects affect the sim-to-real gap in robotics. We investigate what happens if one ignores aspects tied to processes from within the physical camera, e.g., lens distortion, noise, and signal processing; scene effects, e.g., lighting and reflection; and rendering quality. The results of the study demonstrate, quantitatively, that large-scale changes to color, scene, and location have far greater impact than model aspects concerned with local, feature-level artifacts. Moreover, we show that these scene-level aspects can stem from lens distortion and signal processing, particularly when considering white-balance and auto-exposure modeling.
... For the simulated images, the virtual environment was constructed by hand from reference images of the real environment. Included in the baseline camera simulation is a model of a wide-angle lens, based on the polynomial radial model [43], using parameters calibrated from the camera used on the reference vehicle [6]. During data collection with the scaled vehicle, it was noted that the onboard camera did not perform white-balancing, so a color-correction model was added to the camera simulation that unbalanced the colors based on calibration images from the real camera. ...
Preprint
Full-text available
Given the versatility of generative adversarial networks (GANs), we seek to understand the benefits gained from using an existing GAN to enhance simulated images and reduce the sim-to-real gap. We conduct an analysis in the context of simulating robot performance and image-based perception. Specifically, we quantify the GAN's ability to reduce the sim-to-real difference in image perception in robotics. Using semantic segmentation, we analyze the sim-to-real difference in training and testing, using nominal and enhanced simulation of a city environment. As a secondary application, we consider use of the GAN in enhancing an indoor environment. For this application, object detection is used to analyze the enhancement in training and testing. The results presented quantify the reduction in the sim-to-real gap when using the GAN, and illustrate the benefits of its use.
... While this model approximates reasonably low distortions, it is not suitable for thicker lenses with larger fields of view which exhibit large geometric distortions (e.g. fisheye cameras) [49]. To cope with this limitation, appropriate distortion models for large field-of-view cameras have been developed. ...
Preprint
Full-text available
Image editing and compositing have become ubiquitous in entertainment, from digital art to AR and VR experiences. To produce beautiful composites, the camera needs to be geometrically calibrated, which can be tedious and requires a physical calibration target. In place of the traditional multi-images calibration process, we propose to infer the camera calibration parameters such as pitch, roll, field of view, and lens distortion directly from a single image using a deep convolutional neural network. We train this network using automatically generated samples from a large-scale panorama dataset, yielding competitive accuracy in terms of standard l2 error. However, we argue that minimizing such standard error metrics might not be optimal for many applications. In this work, we investigate human sensitivity to inaccuracies in geometric camera calibration. To this end, we conduct a large-scale human perception study where we ask participants to judge the realism of 3D objects composited with correct and biased camera calibration parameters. Based on this study, we develop a new perceptual measure for camera calibration and demonstrate that our deep calibration network outperforms previous single-image based calibration methods both on standard metrics as well as on this novel perceptual measure. Finally, we demonstrate the use of our calibration network for several applications, including virtual object insertion, image retrieval, and compositing. A demonstration of our approach is available at https://lvsn.github.io/deepcalib .
... The most common is the pinhole model (Equation (12)), which is widely applied for monocular and stereo cameras, projecting a world point P to its image coordinate P , where α and β are coefficients of focal length and pixel length, respectively, while c x and c y describe the coordinates of the image's center [153]. Other models-including the polynomial distortion model, the two-plane model, and the catadioptric camera model-are summarized in [154,155]. ...
Article
Full-text available
Although Global Navigation Satellite Systems (GNSSs) generally provide adequate accuracy for outdoor localization, this is not the case for indoor environments, due to signal obstruction. Therefore, a self-contained localization scheme is beneficial under such circumstances. Modern sensors and algorithms endow moving robots with the capability to perceive their environment, and enable the deployment of novel localization schemes, such as odometry, or Simultaneous Localization and Mapping (SLAM). The former focuses on incremental localization, while the latter stores an interpretable map of the environment concurrently. In this context, this paper conducts a comprehensive review of sensor modalities, including Inertial Measurement Units (IMUs), Light Detection and Ranging (LiDAR), radio detection and ranging (radar), and cameras, as well as applications of polymers in these sensors, for indoor odometry. Furthermore, analysis and discussion of the algorithms and the fusion frameworks for pose estimation and odometry with these sensors are performed. Therefore, this paper straightens the pathway of indoor odometry from principle to application. Finally, some future prospects are discussed.
... In order to gather useful data from those images, one needs to calibrate the cameras used, to determine the parameters of the lenses and if needed, compensate for them. Multiple methods have been proposed over the years, most notably by Tsai [159] and later by Zhang [160] and Sturm and Ramalingam [161]. ...
... The projection functions for different lenses are shown in the Table 5 The projection function was already available from a Valeo C++ library. [81,87]. ...
... This model can accurately approximate a moderate amount of radial and tangential distortion. However, the Brown model remains relatively ineffective to represent wide field-of-view cameras which tend to exhibit larger radial distortions (Sturm and Ramalingam, 2011). To tackle this issue, the Kannala-Brandt model (Kannala and Brandt, 2006) proposes a different polynomial form: ...
Article
In this paper, we present MC-Calib, a novel and robust toolbox dedicated to the calibration of complex synchronized multi-camera systems using an arbitrary number of fiducial marker-based patterns. Calibration results are obtained via successive stages of refinement to reliably estimate both the poses of the calibration boards and cameras in the system. Our method is not constrained by the number of cameras, their overlapping field-of-view (FoV), or the number of calibration patterns used. Moreover, neither prior information about the camera system nor the positions of the checkerboards are required. As a result, minimal user interaction is needed to achieve an accurate and robust calibration which makes this toolbox accessible even with limited computer vision expertise. In this work, we put a strong emphasis on the versatility and the robustness of our technique. Specifically, the hierarchical nature of our strategy allows to reliably calibrate complex vision systems even under the presence of noisy measurements. Additionally, we propose a new strategy for best-suited image selection and initial parameters estimation dedicated to non-overlapping FoV cameras. Finally, our calibration toolbox is compatible with both, perspective and fisheye cameras. Our solution has been validated on a large number of real and synthetic sequences including monocular, stereo, multiple overlapping cameras, non-overlapping cameras, and converging camera systems. Project page: https://github.com/rameau-fr/MC-Calib
... This chapter does not contain equations, only geometry -algebraic formulations of some of the material covered can be found in other chapters of this book and another good starting point may be [38]. ...
... These simple geometrical operations can be expressed in a similarly simple algebraic manner (see for instance [38]). ...
... This potentially strong advantage of non-central cameras has to be weighed with care though: if the camera is only "slightly non-central", that is if the camera rays pass all through some small volume in space, then the accuracy of the estimated scale of translation might be poor. More on this issue is explained in section 3.6 of [38]. ...
... This work introduces an evolutionary approach for homography estimation [1] to calibrate fixed-point cameras [2] on the road and estimate vehicle speed using object tracking. ...
... With the known corner position in the camera frame and camera intrinsic parameters, the projection model [42] can be applied to obtain the pixel coordinates of all four corners. ...
Article
Indoor drone or Unmanned Aerial Vehicle (UAV) operations, automated or with pilot control, are an upcoming and exciting subset of drone use cases. Automated indoor flights tighten the requirements of stability and localization accuracy in comparison with the classic outdoor use cases which rely primarily on (RTK) GNSS for localization. In this paper the effect of multiple sensors on 3D indoor position accuracy is investigated using the flexible sensor fusion platform OASE. This evaluation is based on real-life drone flights in an industrial lab with mm-accurate ground truth measurements provided by motion capture cameras, allowing the evaluation of the sensors based on their deviation from the ground truth in 2D and 3D. The sensors under consideration for this research are: IMU, sonar, SLAM camera, ArUco markers and Ultra-Wideband (UWB) positioning with up to 6 anchors. The paper demonstrates that using this setup, the achievable 2D (3D) indoor localization error varies between 4.4 cm and 21 cm (4.9 cm and 67.2 cm) depending on the selected set of sensors. Furthermore, cost/accuracy trade-offs are included to indicate the relative importance of different sensor combinations depending on the (engineering) budget and use case. These lab results were validated in a Proof of Concept deployment of an inventory scanning drone with more than 10 flight hours in a 65 000 m2 warehouse. By combining lab results and real-life deployment experiences, different subsets of sensors are presented as a minimal viable solution for three different indoor use cases considering accuracy and cost: a large drone with little weight- and cost restrictions, one or more medium sized drones, and a swarm of weight and cost restricted nano drones.