Search

In this paper, we propose a method for coarse camera pose computation which is robust to viewing conditions and does not require a detailed model of the scene. This method meets the growing need of easy deployment of robotics or augmented reality applications in any environments, especially those for which no accurate 3D model nor huge amount of ground truth data are available. It exploits the ability of deep learning techniques to reliably detect objects regardless of viewing conditions. Previous works have also shown that abstracting the geometry of a scene of objects by an ellipsoid cloud allows to compute the camera pose accurately enough for various application needs. Though promising, these approaches use the ellipses fitted to the detection bounding boxes as an approximation of the imaged objects. In this paper, we go one step further and propose a learning-based method which detects improved elliptic approximations of objects which are coherent with the 3D ellipsoid in terms of perspective projection. Experiments prove that the accuracy of the computed pose significantly increases thanks to our method and is more robust to the variability of the boundaries of the detection boxes. This is achieved with very little effort in terms of training data acquisition -- a few hundred calibrated images of which only three need manual object annotation. Code and models are released at https://github.com/zinsmatt/3D-Aware-Ellipses-for-Visual-Localization., Comment: Presented at 3DV 2020. Code and models released at https://github.com/zinsmatt/3D-Aware-Ellipses-for-Visual-Localization

International audience; Several separate displays are used by Air Traffic Controllers, such as radar view, flight lists view or paper strips. In order to link the information between these views and keep focus on a subset of flights, controllers are required to do visual operations (eye gaze, analysis of visual properties etc.). This process can be disruptive when traffic increases and when visualizations display large amounts of objects. In this paper we propose the use of animated transitions to replace the visual paths controllers take. We discuss this technique and show results of a predictive evaluation that suggests an improvement in users' performance.

In free navigation applications, any viewpoint to a three-dimensional scene can be synthesized through Depth Image-Based Rendering (DIBR). In this paper we show that XSlit cameras achieve up to 3 dB PSNR gain over conventional pinhole camera arrays when synthesizing virtual views to the scene with DIBR for small camera displacements. XSlit cameras are a type of general linear cameras where the light rays pass through two non-intersecting slits instead of a single point (the optical center of conventional cameras), resulting in a different epipolar geometry and projection equations. Instead of synthesizing virtual views with DIBR from a set of conventional pinhole cameras, e.g. a stereo camera pair, a single XSlit camera exploits the distance between its slits and their relative rotation to obtain disparity, out of which DIBR virtual views can be synthesized. We first present a theoretical study to present the parameters of XSlit cameras, different from pinhole cameras, at the same time making sure that the XSlit camera is physically implementable. We then validate the study with DIBR achieved on synthetic content using perfect depth maps obtained from an in-house modified version of Blender's engine Cycles to simulate XSlit cameras. The virtual view synthesis uses an adapted version of the Reference View Synthesis software used in MPEG, the worldwide standardization committee for media compression. Our experiments show that for the same overall space covered by the studied camera architectures, XSlit cameras often obtain better DIBR view synthesis results, with up to 3 dB PSNR gain., SCOPUS: cp.p, info:eu-repo/semantics/published

International audience; In desktop 3D modeling applications, object manipulation relies on the use of standard transformation widgets. However, these widgets are less efficient on touchscreens due to the low accuracy of touch inputs, and also to the fingertip occlusion issue. In this paper, we propose two unimanual constrained manipulation techniques on TabletPCs. Based on finger identification, our techniques map the index, middle and ring fingers of the dominant hand to the X, Y and Z axes of the coordinate system. After registering the axis-fingers mapping on the screen, users can perform different chording gestures to trigger axis-, plane-constrained or uniform manipulations. We have conducted a user study to examine the learnability and usability of the proposed gestures. The results indicate that these gestures can be memorized quickly after a short training period. Users found some gestures to be easy to use and comfortable while others require additional investigations. These findings permits to give recommendations for designing a unimanual constrained manipulation technique that can be used efficiently in a 3D modeling application on TabletPCs.; Dans les applications de modélisation 3D sur PC, la manipulation avec contraintes d'objets repose sur l’utilisation des widgets de transformation 3D. Cependant, ces widgets sont moins efficaces pour une utilisation sur des écrans tactiles en raison de la faible précision des dispositifs d’entrée et des problèmes d'occultation. Dans ce papier, deux nouvelles techniques uni-manuelles de manipulation 3D sur TabletPC sont proposées. Basées sur l’identification des doigts, nos deux techniques associent l'index, le majeur et l'annulaire de la main dominante aux axes x, y et z du système de coordonnées. Après l’enregistrement de cette association axe-doigts, les utilisateurs peuvent effectuer différents gestes d’accords pour des manipulations avec contraintes sur les axes, sur les plans et des manipulations uniformes. Nous avons mené une étude préliminaire pour examiner l'apprenabilité et l’utilisabilité des gestes proposées. Les résultats indiquent que ces derniers peuvent être mémorisés rapidement. Certains gestes ont été jugés confortables et faciles à utiliser par les utilisateurs alors que d’autres nécessitent davantage d’investigations. Ces résultats permettent de donner des recommandations quant à la conception d’une technique uni-manuelle de manipulation avec contraintes qui peut être utilisée efficacement pour la modélisation 3D sur TabletPC.

In this paper, we report on identifying the underlying factors that contribute to the visual interest in digital photos. A set of 1005 digital photos covering different topics and of different qualities was collected from Flickr. Images were annotated by a pool of diverse participants on a crowdsourcing platform. 12 bipolar ratings were collected for each photo on 7-point semantic differential scale, including dimensions related to interest, emotions and image quality. Every image received 20 annotations from unique participants. The most important appraisals and visual attributes for visual interest in photos was identified. We found that intrinsic pleasantness, arousal, visual quality and coping potential are the most important factors contributing to visual interest in digital photos. We developed a system that automatically detects the important visual attributes from low level visual features and demonstrated their significance in predicting interest at individual level.

National audience; In this paper, we propose SuperVision, a novel interaction technique for controlling distant connect objects in smart-home. Users point an object with their remote control to visualize its state, and select its functionalities. To achieve this goal, 1) we present a novel remote control augmented with a video-projector and a slider; 2) we introduce a visualization allowing users to see through the walls in order to control objects in the line of sight as well as objects in another rooms; 3) we describe applications relying on this interaction technique; Dans ce papier, nous proposons SuperVision, une nouvelle technique d’interaction pour contrôler à distance les "objets connectés" d'une maison intelligente. Les utilisateurs pointent en direction d’un objet à l’aide d’une télécommande pour visualiser son état et sélectionner ses fonctionnalités. Pour ce faire : 1) nous présentons une nouvelle télécommande qui est augmentée d’un pico-projecteur et d’une glissière ; 2) nous introduisons une visualisation permettant à l’utilisateur de "voir" à travers les murs afin de pouvoir contrôler de la même manière les objets visibles et ceux qui ne sont pas dans son champ de vision ; 3) nous décrivons des applications reposant sur cette technique d’interaction.

149-152; International audience; Current window managers are based on the use of overlapping windows, which implies the notion of a window's depth. In order to facilitate the management of windows in desktop environments, recent work has explored their grouping into layers of non-overlapping windows. In this note, we investigate the effect of three visual depth cues on the perception of the depth of such layers: luminosity, blurring and shadow. The paper reports on a controlled experiment that indicates that luminosity helps users to assess the number of layers while blurring and shadow have no effect.

International audience; User interfaces of critical systems, such as air traffic control displays, use graphical objects to code for an ever increasing amount of information. This evolution brings forth concerns about the detection and identification of the displayed objects, in particular for small size objects. First, graphic properties of the interface should include some knowledge about the interplay between colour, shape and size interactions, and the visual perception. Second, the redesign of any interactive system should take into account the particularities of the evolving software and hardware display technology (pixel size and structure, for example) in order to preserve crucial aspects of the initial visual display. The two experiments described in this paper are aiming towards building a more systematic knowledge of graphic properties interactions per se, and their changes as a function of display technologies. More precisely, we examined the effect of the object size, shape and luminosity, as well as its contrast with the background. Results show that object perception is dependent upon its size, its contrast with the background and the overall luminosity of the background. Furthermore, for small size objects, interactions between pixel luminosities and pixel arrangement greatly influence their perception by the human eye.

Segmentation tools have to take into account the technical characteristics of the machine on which they are supported and, of course, the users needs. These tools have to be robust, ergonomic and also need to be fast enough. An ellipse fitting method in ultrasound images is presented in this paper. It makes use the Pascal's theorem and permits the ellipse reconstruction using only 3 points from its contour by the estimation of tangents using the Hough transform. A refining module using genetic algorithm will also be introduced. Preliminary results on different organ segmentation are presented and discussed.

International audience; This paper deals with the extraction of the hedgerow and copse network from hyperspectral images acquired with the Compact Airborne Spectrographic Imager (CASI). The strategy of segmentation integrates several levels of data fusion allowing a decision to be taken concerning the membership of each pixel to the hedgerow and copse network from the large set of original data. The first level leads to quantifying the membership of each pixel to specific features of the network. It includes data fusion based on physical properties, geometric context-dependent fuzzy fusion with an original consistency measure and the geometric fusion of decisions. The second level is a fuzzy fusion of methods allowing the membership of each pixel to the network to be quantified. Finally, the third level involves post-processing the data with a context-dependent fusion of decisions to obtain the final map of the hedgerow and copse network.

Developments in digital signal processing allow sound engineers to restore old recordings. The elementary tools used in the mentioned processes are digital editing of sound supported by the advanced visualization of the processed signal, equalizing filtering and adaptive digital filtering used to diminish continuous noise. Moreover, various methods of the elimination of impulse noises were tried. Some of the above methods were practically applied to the constructed computer sound restoration system presented in the paper. Consequently, the multi-task sound restoration method was proposed and practically employed. Application of some types of procedures to the sound restoration of old recordings will be discussed. The reduction of background noise using digital filtering in the critical bands was investigated. Another procedure for old magnetic tape echoes cancellation was implemented and tested. Application of smoothing filters to the elimination of noises will be also discussed