This paper proposes a robust algorithm that finds a proxy surface from a series of calibrated pictures of an object without assuming any of its reflectance properties. This proxy is optimized to reduce view interpolation errors by globally minimizing the frequency criterion proposed in [1]. The generality of this setting makes robustness particularly difficult to achieve since no model from which to identify outliers or noise is available. Unfortunately, failing to achieve robustness results in unusable proxy for most of the datasets presented. The traditional method of identifying outliers by their discrepency from photoconsistency must somehow be replaced by a global analysis involving all viewpoints. The major contribution of this paper is to meet this requirement by proposing a robust estimation of the minimizer of the frequency criterion as well as a novel framework for merging the multiple depth hypotheses obtained. View interpolation results and proxies are shown for challenging datasets, both Lambertian and not.

   We describe an accurate keypoint detector that is stable under viewpoint change. In this paper, keypoints correspond to actual junctions in the image. The principle of ASN differs fundamentally from other keypoint detectors. At each position in the image and before any detection, it systematically estimates the position of a potential junction from the local gradient field. Keypoints then appear where multiple position estimates are attracted. This approach allows the detector to adapt in shape and size to the image content. One further obtains the area where the keypoint has attracted solutions. Comparative results with other detectors show the improved accuracy and stability with viewpoint change.

   This paper presents a simple linear operator that accurately estimates the parameters of ellipse features. Based on the dual conic model, the operator directly exploits the raw gradient information in the neighborhood of an ellipse’s boundary, thus avoiding the intermediate stage of precisely extracting individual edge points. Moreover, under the dual representation, the dual conic can easily be constrained to a dual ellipse when minimizing the algebraic distance. The new operator is compared to other estimation approaches, including those limited to the center position, in simulation as well as in real situation experiments.

   We present a new geometric calibration method for a structured light system combining a projector with a camera, using a planar target with circular control points. By solely exploiting the mapping between projected conics, the proposed method is strictly geometric and provides unbiased camera to projector correspondences during its application. Such a geometric method does not rely on radiometric calibration. Moreover, the method consistently ensures uniform coverage of the working volume and automatically avoids interference between both the projected and the printed patterns on the calibration target.

   This paper addresses the fundamental problem of automatically discovering an unknown moving deformable object in a monocular video sequence. No prior model of the object is used; it is only assumed that the object is composed of a set of apparently rigid parts that are not necessarily visible simultaneously, making it possible to circumvent the typical constraint of model initialization. A set of rigid parts describing the object is incrementally extracted in a modeling-tracking loop with reinforced memory. In this framework, low-level segmentation is considered as a necessary but non reliable process that helps initiating hypotheses. Motion-based layer segmentation from feature points and edges is applied only when and where no modeled parts can be tracked. Using the quantity of motion measure, it is further shown how to deal with temporal scale. The interest for this approach in applications such as human tracking is demonstrated for a set of various sequences including a rapidly evolving shape.

Résumé en anglais

   The median filter is one of the basic building blocks in many image processing situations. However, its use has long been hampered by its algorithmic complexity of O(r) in the kernel radius. With the trend toward larger images and proportionally larger filter kernels, the need for a more efficient median filtering algorithm becomes pressing. In this correspondence, a new, simple yet much faster algorithm exhibiting O(1) runtime complexity is described and analyzed. It is compared and benchmarked against previous algorithms. Extensions to higherdimensional or higher-precision data and an approximation to a circular kernel are presented as well.

   This paper adopts a sampling perspective to surface light field modeling. This perspective eliminates the need of us- ing the actual object surface in the surface light field defini- tion. Instead, the surface ought to provide only a parame- terization of the surface light field function that specifically reduces aliasing artifacts visible at rendering. To find that surface, we propose a new criterion that aims at optimiz- ing the smoothness of the angular distribution of the light rays emanating from each point on the surface. The main advantage of this approach is to be independent of any spe- cific reflectance model. The proposed criterion is compared to widely used criteria found in multi-view stereo and its ef- fectiveness is validated for modeling the appearance of ob- jects having various unknown reflectance properties using calibrated images alone.

   This paper presents the full proof of concept of a system for capturing the light field of an object. It is based on a single high resolution camera that is moved all around the object on a cable-driven end-effector. The main advantages of this system are its scalability and low interference with scene lighting. The camera is accurately positioned along hemispheric trajectories by observing target features. From the set of gathered images, the visual hull is extracted and can be used as an approximate geometry for mapping a surface light field. The paper describes the acquisition system as well as the modeling process. The ability of the system to produce models is validated with four different objects whose sizes range from 20 cm to 3 m.

   This paper presents a system for 3D reconstruction using a camera combined with an inertial sensor. The system mainly exploits the orientation obtained from the inertial sensor in order to accelerate and improve the matching process between wide baseline images. The orientation further contributes to incremental 3D reconstruction of a set of feature points from linear equation systems. The processing can be performed online while using consecutive groups of three images overlapping each other. Classic or incremental bundle adjustment is applied to improve the quality of the model. Test validation has been performed on object and camera centric sequences.

   A few 3D interactive modeling systems have been developed recently. Such systems must cope with a high flow of input measurements during the entire acquisition period. Therefore, the reconstruction and rendering algorithms used must all run online. However, compression algorithms are still run offline as postprocessing. In order to develop a fully interactive modeling framework, this paper presents an online compression algorithm where the system automatically adjusts the level of detail according to the user behavior. The proposed method can reduce peak memory consumption by more than 50% during the acquisition of a typical model and the final result is comparable to offline compression. Furthermore, the results obtained show that a local acquisition approach must be prioritized.

   In image-based light field rendering, many efforts have been made to improve the efficiency of the modeling. This efficiency pertains to the photorealism and compression of a given model. We propose a new way to improve these two aspects. We show that it is better to sample the plenoptic function on a surface that minimizes the frequency content of all its lumispheres. We also demonstrate an additional constraint based on the visual hull that further guides the sampling of the plenoptic function around an object. We propose a corresponding algorithm that relies on images alone and only suppose that the modeled object is opaque. This algorithm is then validated on both real and synthetic data sets.

   3D interactive modeling from range data aims at simultaneously producing and visualizing the surface model of an object while data is collected. The current research challenge is producing the final result in real-time. Using a recently proposed framework, a surface model is built in a volumetric structure encoding a vector field in the neighborhood of the object surface. In this paper, it is shown that the framework allows one to locally control the model resolution during acquisition. Using ray tracing, efficient visualization approaches of the multiresolution vector field are described and compared. More precisely, it is shown that volume traversal can be optimized while preventing holes and reducing aliasing in the rendered image.

   Interactive 3D modeling is the process of building a 3D model of an object or a scene in real-time while the 3D (range) data is acquired. This is possible only if the computational complexity of all involved algorithms is linear with respect to the amount of data. We propose a new framework for 3D modeling where a complete modeling chain meets with this requirement. The framework is based on the use of vector elds as an implicit surface representation. Each modeling step, registration, surface reconstruction, geometric fusion, compression and visualization is solved and explained using the vector elds without any intermediate representations. The proposed framework allows model reconstruction from any type of 3D data, surface patches, curves, unorganized sets of points or a combination of these.

   Traditional approaches for surface reconstruction from range data require that the input data be either range images or unorganized sets of points. With these methods, range data acquired along curvilinear patterns cannot be used for surface reconstruction unless constraints are imposed on the shape of the patterns or on sensor displacement. This paper presents a novel approach for reconstructing a surface from a set of arbitrary, unorganized and intersecting curves. A strategy for updating the reconstructed surface during data acquisition is described as well. Curves are accumulated in a volumetric structure in which a vector field is built and updated. The information that is needed for efficient curve registration is also directly available in the vector field. The proposed modeling approach combines surface reconstruction and curve registration into a unified procedure. The algorithm implementing the approach is of linear complexity with respect to the number of input curves and makes it suitable for interactive modeling. Simulated data based on a set of six curvilinear patterns as well as data acquired with a range sensor are used to illustrate the various steps of the algorithm.

   This paper proposes two indicators for predicting the quality of camera model parameters from a set of input images. The first indicator is based on the acutance. It can quickly indicate static or motion blur during image capture and correlates well with the 3D reconstruction error when using stereo cameras. The second indicator provides the overall distribution of control points in a set of input images. Although the importance of covering the entire field of view is verified, the spatial distribution of control points need not be uniform. The analysis is supported by experiments with mono and stereo cameras calibrated using both a low cost planar and a high quality 3D target.

   This paper proposes an approach for surface reconstruction of free-form rigid objects from an arbitrary set of intersecting range curves. A strategy for updating the reconstructed surface during data acquisition is described as well. Geometric and color information is accumulated in a volumetric structure in which a vector field is built and updated. Moreover, the information that is needed for efficient curve registration is directly available in this vector field. This leads to a unified modeling approach combining surface reconstruction and curve registration. The algorithm implementing the approach is of linear complexity with respect to the number of input curves and makes it suitable for interactive modeling. A compression scheme based on a multiresolution decomposition of vector fields is introduced as well. Simulated data from a set of curvilinear patterns as well as data acquired with a hand-held range sensor are used to validate the approach.