The problem of using light stripe projection (LSP) for reconstructing the 3D surface of several objects positioned at different distances from the camera is addressed in this paper. It is shown using LSP for 3D surface reconstruction of objects at different depths can be significantly improved if adaptive edge detection filter is added. The filter is designed as a local dynamic hysteresis thresholding value generator. It adapts knowledge of the wavelet coefficients in a small neighborhood on the row profile and generates the local hysteresis threshold values to detect a meaningful and useful edge. It is then shown how the filter may be designed to recover missing LSP features by using common edge detection technique and how it increases the accuracy and reliability of 3D surface reconstruction. The adaptive edge detection filter is illustrated by the presentation of experimental results obtained using the methods described in the paper. A test outlines the differences between LSP edges detected with the adaptive edge detection filter and edges detected without the filter.

A new pixel structure for the demodulation of intensity modulated light waves is presented. The integration of such pixels in line and area array sensors finds application in time-of-flight three-dimensional imaging. In 3D range imaging an illumination module sends a modulated optical signal to a target, where it is reflected back to the sensor. The phase shift of the reflected signal compared to the emitted signal is proportional to the distance to one point of the target. The detection and demodulation of the signal is performed by a new pixel structure named drift field pixel. The sampling process is based on the fast separation of photogenerated charge due to lateral electrical fields below a high-resistive transparent poly-Si photogate. The dominant charge transfer phenomenon of drift, instead of diffusion as in conventional CCD pixels, allows much higher modulation frequencies of up to 1 GHz and a much higher ultimate distance accuracy as a consequence. First measurements performed with a prototype pixel array of 3x3 pixels in a 0.8 micron technology confirm the suitability of the pixels for applications in the field of 3D-imaging. Depth accuracies in the sub centimeter range have already been achieved.

Optical Coherence Tomography (OCT) is an optical imaging technique allowing the acquisition of three-dimensional images with micrometer resolution. It is very well suited to cross-sectional imaging of highly scattering materials, such as most biomedical tissues. A novel custom image sensor based on smart pixels dedicated to parallel OCT (pOCT) is presented. Massively parallel detection and signal processing enables a significant increase in the 3D frame rate and a reduction of the mechanical complexity of the complete setup compared to conventional point-scanning OCT. This renders the parallel OCT technique particularly advantageous for high-speed applications in industrial and biomedical domains while also reducing overall system costs. The sensor architecture presented in this article overcomes the main challenges for OCT using parallel detection such as data rate, power consumption, circuit size, and optical sensitivity. Each pixel of the pOCT sensor contains a low-power signal demodulation circuit allowing the simultaneous detection of the envelope and the phase information of the optical interferometry signal. An automatic photocurrent offset-compensation circuit, a synchronous sampling stage, programmable time averaging, and random pixel accessing are also incorporated at the pixel level. The low-power demodulation principle chosen as well as alternative implementations are discussed. The characterization results of the sensor exhibit a sensitivity of at least 74 dB, which is within 4 dB of the theoretical limit of a shot-noise limited OCT system. Real-time high-resolution three-dimensional tomographic imaging is demonstrated along with corresponding performance measurements.

Laser scanners are often used in geometric reverse engineering to generate a CAD drawing out of an existing physical part. This often requires careful path planning to ensure the correct stand-off distance of the sensor, to prevent any collisions of the sensor with the target object and to prevent any scan occlusions in the resulting data. A dedicated system was constructed to collect data using a 3-axis cylindrical/prismatic device. The benefit of this scanning system is the ability to scan completely around the target object, including the bottom of the object. A stripe-type laser diode is used along with a CCD camera. The target object sits on a transparent plexi-glass table, which can be rotated 360 degrees. Both the laser and the CCD are mounted on an arm, such that they can be traversed along a linear path, as well as being rotated 360° around the table. This permits the device to scan in either spherical or cylindrical paths with 360° rotation around the target object. This system allows for the quick scanning of any object while minimizing the number of occlusions in the resulting scan data.

Normal stereovision system requires two or more cameras to capture different views of the same scene. One category of technique called single-lens stereovision attracted many researchers interest because of its significant advantages over the normal stereovision setup including compactness, low cost, less system parameters and ease of calibration, etc. In this paper we present some new understanding of a single-lens stereovision system using a biprism (2F filter). Image captured by the real camera with a biprism placed before its lens, is divided into two equal halves. Each half-image is assumed to be captured by one virtual camera. Two related but different approaches of understanding and modeling such a system are introduced: one is based on camera calibration technique and another is based on geometrical analysis. The latter approach provides an interesting way of understanding this system. It does not require complex calibration, and one field point test is sufficient to determine the system once the system is built and pin-hole camera model is used. Thus, great effort on setup and calibration will be saved compared to normal binocular stereovision system. The approach based on geometrical analysis provides a relatively simpler and sufficiently accurate way of building a close range stereovision system.

The chromatic confocal approach enables the parallelization of the complete depth-scan of confocal topography measurements. Therefore, mechanical movement can be reduced or completely avoided and the measurement times shortened. Chromatic confocal point sensors are already commercially available but they need lateral scanning in x- and y-direction in order to measure surface topographies. We achieved a further parallelization in the x-direction by realizing a chromatic confocal line sensor using a line focus and a spectrometer. In a second setup, we realized an area measuring chromatic confocal microscope, which is capable of one-shot measurements without any mechanical scanning. The depth resolution of this setup can be improved by measuring in a small number of different heights. Additional information about the color distribution of the object is gained.

A binocular robot vision system having an autonomously moving active viewpoint is proposed. By using this active viewpoint, the system constructs a correspondence between the images of a feature points on the right and left retinas easily and calculates the spatial coordinates of the feature points. The system incorporates two intelligent functions for enlarging the measuring region and increasing the accuracy of measurement. The first intelligent function is an autonomous movement of the viewpoint and the second is weighting process for measured points. These functions work when the system can recognize the solid body. As the first steps of the development of the recognition of a solid body, we incorporate these functions into the system to detect straight lines in an image. To detect lines we use Hough transform. The system searches a region surrounded by 4 straight lines. Then the system recognizes the region as a quadrangle. The system constructs a correspondence between the quadrangles in the right and left images. By the use of the result of the constructed correspondence, the system calculates the spatial coordinates of an object. An experiment shows the effect of the line detection using Hough transform, the recognition of the surface of the object and the calculation of the spatial coordinates of the object.

This paper presents a purely CMOS Active Pixel Sensor (APS) capable of time-of-flight (TOF) range imaging. The TOF sensor introduced is used to calculate range in the charge domain. To obtain high speed and highly efficient charge transfer that is important for TOF range imaging, a high gain inverting amplifier and two capacitors connected alternatively to the feedback path are used. The high speed and highly efficient charge transfer using a high gain inverting amplifier in a negative feedback loop enables the TOF range imaging to be based on standard CMOS technology. Moreover, CMOS based amplifier circuits have matured and are relatively easy to design. The analysis of the APS together with simulation results suggests that the proposed technique can achieve a sufficient range resolution of millimeter to centimeter depending on the maximum measured range.

This paper presents a simple method to calibrate the intrinsic parameters of zoom-lens digital cameras. This method combines the classical calibration algorithm using a planar pattern and the Exchangeable Image File Format (EXIF) metadata of image files captured by digital cameras. The EXIF metadata records many information about the camera’s setting such as the focal length of zooming lens. So we can use the focal length from EXIF to know the zoom lens setting. Firstly, a pre-calibration should be done to know the relationship between zoom lens settings and the intrinsic camera parameters. We take some sample lens settings from the minimum focal length to the maximum one by changing the lens zooming positions, and perform the mono focal calibration for each lens setting configuration. Then we get the coefficients of the polynomial function through curve fitting. After that we can get the intrinsic parameters correspond with the zoom lens setting of new image files shoot by this digital camera. Our experiments show the proposed method can provide accurate intrinsic camera parameters for all the lens settings continuously.

Usually in an image no real information about the scene’s depth (in terms of absolute distance) is available. In this paper, a method that extracts real depth measures is developed. This approach starts considering a region located in the center of the depth map. This region can be positioned, interactively, in any part of the depth map in order to measure the real distance of every object inside the scene. The histogram local maxima of this region are determined. Among these values the biggest, that represents the gray-level of the most considerable object, is chosen. This gray-level is used in an exponential mapping function that converts, using the input camera settings, the depth map gray-levels into real measures. Experiments over a large dataset of images show good performances in terms of accuracy and reliability.

This paper presents a novel and fully automatic technique to estimate depth information from a single input image. The proposed method is based on a new image classification technique able to classify digital images (also in Bayer pattern format) as indoor, outdoor with geometric elements or outdoor without geometric elements. Using the information collected in the classification step a suitable depth map is estimated. The proposed technique is fully unsupervised and is able to generate depth map from a single view of the scene, requiring low computational resources.

In this paper, a topology-based strategy for 3D reconstruction of complicated buildings from stereo image pair is put forward. It comes from our investigation on the applicability of topology analysis and a strongly topology-driven process that combines different levels of geometrical description with different levels of topological abstraction. The authors emphasize the topology-based strategy on different levels of geometrical description: Firstly a topology-based 3D data model is presented in which the topological relationships within a building or between geometrical objects are described implicitly or explicitly. Secondly based on description of vertexes level, interested vertexes are collected from stereo image pair and saturated attribute of each interior vertex is defined, furthermore an adjacency table is defined to store the connection attributes of verges automatically. Thirdly surfaces are looked on as polygons with closed verges on the basis of bi-directional querying of the adjacency table. Finally complicated buildings are described as graphs with interior and exterior topological attributes. Based on the strategy mentioned above, a software platform for 3D reconstruction of complicated buildings is built up. The efficiency of suggested method is examined through practical experiments.

Fall protection harnesses are commonly used to reduce the number and severity of injuries. Increasing the efficiency of harness design requires the size and shape variation of the user population to be assessed as detailed and as accurately as possible. In light of the unsatisfactory performance of traditional anthropometry with respect to such assessments, we propose the use of 3D laser surface scans of whole bodies and the statistical analysis of elliptic Fourier coefficients. Ninety-eight male and female adults were scanned. Key features of each torso were extracted as a 3D curve along front, back and the thighs. A 3D extension of Elliptic Fourier analysis4 was used to quantify their shape through multivariate statistics. Shape change as a function of size (allometry) was predicted by regressing the coefficients onto stature, weight and hip circumference. Upper and lower limits of torso shape variation were determined and can be used to redefine the design of the harness that will fit most individual body shapes. Observed allometric changes are used for adjustments to the harness shape in each size. Finally, the estimated outline data were used as templates for a free-form deformation of the complete torso surface using NURBS models (non-uniform rational B-splines).

Three-dimensional (3D) object reconstruction from range images plays an important role in many research and application fields, including computer vision, reverse engineering, computer graphics, and CAD/CAM. Since data integration is a fundamental step in object reconstruction, a great number of research efforts have been made on that. In this paper, a novel integration algorithm is presented. Firstly, the input data (registered data) which contains overlapping data is represented by kd-tree structure. Then, three theorems are provided together with the usage of nearest neighbor searching to identify and eliminate the overlapping data. The method manipulates the registered data directly without preprocessing work, therefore, provides an efficient and straightforward way to remove the redundant data. This is different from the traditional methods which need to mesh the input data or build an implicit surface function before integration. To reduce the data size and obtain a reasonable density distribution, a reliable resampling method called ball travel based resampling is also developed. The experimental results demonstrate the efficiency of the proposed algorithm.

Lots of techniques exist to generate a 3D model given a limited set of photographs. Extracting depth information using stereoscopic correspondences generally outputs a large number of depth maps to be merged. Generating a dense representation of the object (shape-from-shading, level-set minimization) can be time-consuming. Shape-from-silhouettes and hybrid techniques, carving an approximate model given consistency constraints, suffer from a lack of scalability. In the context of the GANTOM project, we developed a system based on an automatic turntable, high-resolution digital cameras (up to 4Mpixels) and accurate encoders to acquire the data. This allows us to perform a very fine calibration of the acquisition process, i.e. less than 0.5mm. In this paper, we present the GANTOM software system that can be subdivided into the following elements: - Depth map extraction using stereoscopic correspondence. Different search spaces are compared. - Depth map fusion using a voxel-based algorithm, called shape-from-depth, which is a generalization of shape-from-silhouettes. - Conversion from a volumetric model to a polyhedral model. - Texture extraction using the polyhedral model and the photographs as inputs.

One of the most powerful applications of the World Wide Web (WWW) is the storage and distributuion of multimedia, integrating text, images, sound, videos and hyperlinks. In cultural heritage this is of particular interest, because best methods to convey a complex knowledge in the field of cultural heritage, to experts and non-experts, are the visual representation and visual interaction. In this work we propose 3D acquisition and digitizing techniques for the virtualized reality of small cultural heritage objects (virtual gallery). The system used for creating 3D shape is based on the conoscopic holograph. This technique is a non-contact three-dimensional measuring technique that makes possible to produce holograms, even with incoherent light, with fringe periods that can be measured precisely to determine the exact distance to the point measured. It is suitable to obtain 3D profile with high resolution also on surface with unevenness reflectivity (this situation is usual on the surface of the cultural heritage objects). By conoscopic holography, high-resolution 3D model can be obtained. Howver, accurate representation and high-quality display are fundamental requirements to avoid misinterpretation of the data. Therefore, virtual gallery can be obtained through a procedure involving 3D acquisition, 3D model and visualization.

In this paper, we present a 3D object reconstruction system that recovers 3D models of general objects from video. We assume the video of the object is captured from multiple viewpoints. The proposed system is composed of the following components: feature trajectory extraction, 3D structure from motion, surface reconstruction, and texture computation. In the feature trajectory extraction, we compute dense optical flow fields between adjacent frames and aggregate them at the interest points to obtain reliable feature trajectories. In the next structure from motion stage, we develop a robust algorithm to recover the dense 3D structures from several viewpoints for uncalibrated image sequences. For the surface reconstruction from the recovered 3D data points, we develop a new cluster-based radial-basis-function (RBF) algorithm, which overcomes the extensive computational cost limit in a divide-and-conquer manner. For the last texture computation process, we combine multi-view images to form the texture map of the 3D object model. Finally, experimental results are given to show the performance or the proposed 3D reconstruction system.

This paper proposes an adaptive regularized noise smoothing algorithm for range images using the area decreasing flow method, which can preserve meaningful edges during the smoothing process. Adaptation is incorporated by adjusting the regularization parameter according to the results of surface curvature analysis. In general, range data includes mixed noise such as Gaussian or impulsive noise. Although non-adaptive version of regularized noise smoothing algorithm can easily reduce Gaussian noise, impulsive noise caused by random fluctuation of the sensor acquisition is not easy to be removed from observed range data. It is also difficult to remove noise near edge using the existing adaptive regularization algorithms. In order to cope with the problem, the second smoothness constraint is additionally incorporated into the existing regularization algorithm, which minimizes the difference between the median filtered data and the estimated data. As a result, the proposed algorithm can effectively remove the noise of dense range data while meaningful edge is well-preserved.