Although laser range sensors based on sequential scanning can provide accurate measurements in stable operation, the recovered surface geometry becomes noisy and distorted when sensors are hand-held. To compensate for camera motion, some currently existing prototypes integrate a positioning device. Unfortunately, these may not be accurate and fast enough. To circumvent this problem, a method that can compensate for motion distortion is proposed. The principle consists in using the measured shape geometry as a reference frame in 3-D space. The method is based on the collection of a redundant set of crossing profiles. Each surface profile is measured in a very short period of time such that distortion of the profile is negligible. It is assumed that the perturbation error due to motion, affects inter-profile positioning only. Then, the set of rigid crossing profiles are fitted together by moving them such as to minimize the profile intersection spacings. Experiments show that errors in the geometry can be reduced to the order of magnitude of the sensor error. The method can be integrated in the design of a hand-held sensor or as a complementary post-processing stage for improving measurement accuracy when using a sensor positioning device.
The stereo analysis method is similar to the human visual system. Due to the way our eyes are positioned and controlled, our brains usually receive similar images of a scene taken from nearby points of the same horizontal level. Stereo tries to imitate this principle by computing the distance of objects, 'their depth,' from two images of two cameras using the triangulation principle. Points of imaged objects are mapped in different locations in the two stereo images. These points need to be identified in both images. The correspondences are established by correlating windows of the left and right image and finding a maximum. A central problem in stereo matching using correlation techniques lies in selecting the size of the search window. Small windows contain only a small number of data points, and thus are very sensitive to noise and therefore result in false matches. Whereas large search windows contain data from two or more different objects or surfaces, thus the estimated disparity is not accurate due to different projective distortions in the left and the right image. The new method introduces a continuous scale parameter for the matching process. It allows the adaption of the scale for every individual region and overcomes the drawbacks of fixed window sizes which is impressively demonstrated by the experimental results.
The ultimate goal for future telecommunication is highly effective inter-personal information exchange. The effectiveness of telecommunication is greatly enhanced by 3-D telepresence. This requires that visual information is presented in such a way that the viewer is under the impression of actually being physically close to the party with whom the communication takes place. One way to achieve a natural 3-D impression is to encode image sequences using 3-D model objects and animate them again by computer graphic means regarding the observers eye positions. This concept will use a parametric 3-D scene description in order to model a scene. The parameters of the model objects will be estimated from trinocular input image sequences by means of image analysis. This paper starts with an overview on the European ACTS project PANORAMA, in which the above mentioned concept will be realized and evaluated. In the main part the shape initialization of physical objects from a multiview image sequence will be discussed. For this the range information given by three disparity maps from different stereo views is backprojected into 3-D space. The resulting cloud of 3-D points is then approximated by a flexible triangular net by using a technique named discrete smooth interpolation. The discrete smooth interpolation is a particular surface interpolation technique, which is solved by an iterative approach. It allows to generate a surface, defined as a wireframe mesh, that fits (or interpolates) a given set of 3D points by observing, at the same time, some given constraints about the surface characteristics, like roughness, behavior at the boundaries, etc. The finally presented results show the capabilities of this approach in video communication.
A new, simple technique is proposed to recover 3D shape from 2D images. For this purpose, an opposite camera geometry is introduced to decrease stereo matching calculation time. In order to get 2D image data, a symmetric camera geometry is adopted; that is, two cameras are located oppositely and their lens centers are on a straight line. Then, the orthographic projection images are make the same silhouettes of the object. However in the case of the perspective projection, the two shapes are different and the difference is dependent on the difference of the distances from each camera. So, when the same point is observed on the edge of the silhouettes, we can find the corresponding point at the same direction in lens center coordinate system. From this situation, it is easy to solve the stereo matching problem. Moreover, if we know the camera's position, the 3D position of such object's point is determined from each radius in polar, because the radii of the two images are simply proportional to the camera distances. An experiment for the stuffed rabbit demonstrates the validity of this method.
This paper presents a system that creates realistic 3D representations of real indoor environments using laser and video. The principle novelty of the system is in devising an automatic procedure for planning successive positions of the laser device within a building and integrating the data acquired into a coherent overall model. This is achieved as follows: (1) detection of the occlusions present in the range data (normally achieved by analyzing the continuity of the depth map); (2) determination of the set of possible capture points from where occlusions can be resolved, (3) selection of the capture points that optimize a number of criteria (e.g., number of acquisition sessions, practicability of moving to a capture point, better acquisition conditions). To generate realistic models, a spatial camera is used to extract information from the environment; spatial relationships and visual appearance. The spatial camera includes sensors and actuators, in particular a laser range finder and a TV camera, all mounted on a mobile platform. The system is necessarily mobile since resolving occlusions requires the acquisition of spatial data from multiple capture points.
The development of a family of partial and whole body scanners provides a complete technology for fully three-dimensional and contact-free scans on human bodies or other living objects within seconds. This paper gives insight into the design and the functional principles of the whole body scanner VIRO 3D operating on the basis of the laser split-beam method. The arrangement of up to 24 camera/laser combinations, thus dividing the area into different camera fields and an all- around sensor configuration travelling in vertical direction allow the complete 360-degree-scan of an object within 6 - 20 seconds. Due to a special calibration process the different sensors are matched and the measured data are combined. Up to 10 million 3D measuring points with a resolution of approximately 1 mm are processed in all coordinate axes to generate a 3D model. By means of high-performance processors in combination with real-time image processing chips the image data from almost any number of sensors can be recorded and evaluated synchronously in video real-time. VIRO 3D scanning systems have already been successfully implemented in various applications and will open up new perspectives in different other fields, ranging from industry, orthopaedic medicine, plastic surgery to art and photography.
This paper presents a recently developed custom footwear system, which integrates 3D digitization technology, range image fusion techniques, a 3D graphical environment for corrective actions, parametric curved surface representation and computer numerical control (CNC) machining. In this system, a support designed with the help of biomechanics experts can stabilize the foot in a correct and neutral position. The foot surface is then captured by a 3D camera using active ranging techniques. A software using a library of documented foot pathologies suggests corrective actions on the orthosis. Three kinds of deformations can be achieved. The first method uses previously scanned pad surfaces by our 3D scanner, which can be easily mapped onto the foot surface to locally modify the surface shape. The second kind of deformation is construction of B-Spline surfaces by manipulating control points and modifying knot vectors in a 3D graphical environment to build desired deformation. The last one is a manual electronic 3D pen, which may be of different shapes and sizes, and has an adjustable 'pressure' information. All applied deformations should respect a G1 surface continuity, which ensure that the surface can accustom a foot. Once the surface modification process is completed, the resulting data is sent to manufacturing software for CNC machining.
Reliable 3D wholebody scanners which output digitized 3D images of a complete human body are now commercially available. This paper describes a software package, called 3DM, being developed by researchers at Clemson University and which manipulates and extracts measurements from such images. The focus of this paper is on tilted planes, a 3DM tool which allows a user to define a plane through a scanned image, tilt it in any direction, and effectively define three disjoint regions on the image: the points on the plane and the points on either side of the plane. With tilted planes, the user can accurately take measurements required in applications such as apparel manufacturing. The user can manually segment the body rather precisely. Tilted planes assist the user in analyzing the form of the body and classifying the body in terms of body shape. Finally, titled planes allow the user to eliminate extraneous and unwanted points often generated by a 3D scanner. This paper describes the user interface for tilted planes, the equations defining the plane as the user moves it through the scanned image, an overview of the algorithms, and the interaction of the tilted plane feature with other tools in 3DM.
The Cyberware WB4 whole body scanner generates a high- resolution data set of the outer surface of the human body. The acquisition of anthropometric data from this data set is important for the development of custom sizing for the apparel industry. Software for locating anthropometric landmarks from a cloud of more than 200,000 three-dimensional data points, captured from a human subject, is presented. The first phase of identification is to locate externally placed fudicials on the human body using luminance information captured at scan time. The fudicials are then autonomously labeled and categorized according to their general position and anthropometric significance in the scan. Once registration of the landmarks is complete, body measurements may be extracted for apparel sizing.
For 3D digitizers to be useful data collection tools in scientific and human factors engineering applications, the models created from scan data must match the original object very closely. Factors such as ambient light, characteristics of the object's surface, and object movement, among others can affect the quality of the image produced by any 3D digitizing system. Recently, Cyberware has developed a whole body digitizer for collecting data on human size and shape. With a digitizing time of about 15 seconds, the effect subject movement, or sway, on model fidelity is an important issue to be addressed. The effect of sway is best measured by comparing the dimensions of an object of known geometry to the model of the same object captured by the digitizer. Since it is difficult to know the geometry of a human body accurately, it was decided to compare an object of simple geometry to its digitized counterpart. Preliminary analysis showed that a single cardboard tube would provide the best artifact for detecting sway. A tube was attached to the subjects using supports that allowed the cylinder to stand away from the body. The stand-off was necessary to minimize occluded areas. Multiple scans were taken of 1 subject and the cylinder extracted from the images. Comparison of the actual cylinder dimensions to those extracted from the whole body images found the effect of sway to be minimal. This follows earlier findings that anthropometric dimensions extracted from whole body scans are very close to the same dimensions measured using standard manual methods. Recommendations for subject preparation and stabilization are discussed.
A system for three-dimensional, non-destructive acquisition of the structure of plant root systems is described. The plants are grown in a transparent medium (a 'gel pack') and are then placed on a rotating stage. The stage is rotated in 5-degree increments while images are captured using either traditional photography or a CCD camera. The individual images are then used as input to a tomographic (backprojection) algorithm to recover the original volumetric data. This reconstructed volume is then used as input to a 3D-reconstruction system. The software performs segmentation and mesh generation to derive a tessellated mesh of the root structure. This mesh can then be visualized using computer graphics, or used to derive measurements of root thickness and length. For initial validation studies, a wire model of known length and gauge was used as a calibration sample. The use of the transparent gel- pack media, together with the gel tomography software, allows the plant biologist a method for non-destructive visualization and measurement of root structure that has previously been unattainable.
This paper describes a sensor that employs three laser beams in order to measure the 3-D distress characteristics of road surfaces at highway speed. The 3-D feature extraction is done by means of a combination of two measuring techniques: triangulation and defocusing. In order to generate a pattern of transverse profiles spaced by 11 cm at a speed of about 20 m/s, three transverse profiles are acquired in a single frame, at a frame rate of 60 Hz. This approach led to the three laser beam solution and the use of a standard CCD camera as detector. Subject analysis emphasized as a 'must have': a wide field of view, low blurring, a more even distribution of the laser line intensity, a low power consumption along with simplicity and low cost. The use of a modified reversed prism provided a more uniform distribution, even with a single laser for each projected line. High power laser diodes were needed in order to operate at 1/10,000s shutter speed. An accurate sensor calibration provides 256 points of z (range), x (position) and I (intensity) for each profile. Calibration data as well as range and intensity of measured profiles are provided. Sensor's accuracy is also observed.
Optical 3D sensors are used as tools for reverse engineering: First the shape of an object is digitized by acquisition of multiple range images from different view points. Then the range images are registered and the data is turned into a CAD description, e.g. tensor product surfaces, by surface modeling software. For many applications however it is sufficient to generate a polyhedral surface. We present a nearly automatic procedure covering the complete task of data acquisition, calibration, surface registration and surface reconstruction using a mesh of triangles. A couple of measurements, such as teeth, works of art and cutting tools are shown.
We present an image processing method which can be useful for objects analysis dealing with life and earth sciences. We use a non-contact optical sensor based upon laser triangulation to determine depth of 3D surfaces of the object under consideration. The 3D surface constitutes an image which can be characterized by multiresolution analysis. We applied this system to the study of bivalve shell surface (Unio sp. Recent Atlantic, Holocene) to determine the influence of the environment on its growth rate. Our algorithm is based upon the wavelet transform.
We offer to use the 3D surface profile real-time measurement using phase-shifted interference fringe projection technique for the cranioficial identification. Our system realizes the profile measurement by projecting interference fringe pattern on the object surface and by observing the deformed fringe pattern at the direction different from the projection. Fringes are formed by a Michelson interferometer with one mirror mounted on a piezoelectric translator. Four steps self- calibration phase-shift method was used.
The results of development of noncontact optico-electronic real time system for integral dimensional inspection of atomic reactor parts, as hollow cylinders made of soft alloy, are given. This cheap, reliable and robust system for the industrial operation is based on shadow method for measuring the outside diameter and straightness deviation as well as on the triangulation method for measuring the inside diameter. For the shadow method the main sources of errors are caused by the influence of volumetric properties of 3D objects on the structure of shadow image, the inaccuracy of approximation of object's edge position caused by linear photodiode array as well as the aberration of optical system. By choosing type of the optical system and using look-up table algorithm for aberration correction as well as the cubic approximation of object's edge profile under its reconstruction allows to reduce the maximal total error up to 2 micrometer when the speed of measurement is 500 Hz. The optimization of triangulation system parameters under operation with technical (machined) surface objects in a real hard workshop conditions (high temperature range, dust) was made. The using hardware signal processor for median algorithm data processing is optimal as to the accuracy and frequency response of measurement. The developed system allows inspection articles with outside diameter up to 50 mm, inside diameter 43 mm, length -- up to 150 mm, straightness deviation equal to 0,3 mm with errors 0.014 mm for diameters, 0.040 mm for length and 0.020 mm for straightness deviation. At present the system is operating on the technological conveyer line for real time inspection of atomic reactor details (loading and unloading are made by manipulator) on the nuclear enterprise.
Geomagic WrapTM is a commercially available software for reconstructing shapes and surfaces from 3D scanning data. The data can be any arbitrary finite point set in 3D, and there are no requirements on local density or organization in slices, etc. The software contains components for surface reconstruction, improvement, and analysis, and it supports a variety of input and output formats that make it compatible with scanning hardware and CAD and graphics software.
In this paper, a new real-time 3D reconstruction system is proposed. It's based on a content addressable memory (CAM)- based highly parallel 3D voting algorithm from multiple cameras. By 3D voting we mean a counting the number of back- projection lines passing for each voxel. For reducing the memory size of 3D voxel space and highly parallel processing, our highly parallel processing board using CAM is very suitable. Because coordinates calculation for back-projection lines and coordinates searching for counting back-projection line passing can be carried out in a massively parallel manner, a CAM provides greater parallelism within a limited chip area. Thus, a compact and high level of parallelism can be achieved. Experimental results indicate that a real-time 3D reconstruction of 100 feature points in a 128 cube space from four 128 square images is possible. Therefore, a CAM-based 3D reconstruction system can be considered a promising candidate in the realization of a real-time motion capturing system.
In order to digitize the whole surface of a three-dimensional object by means of an optical range sensor, usually multiple range images are acquired from different viewpoints and merged into a single surface description. The simplest and most accurate way is to generate a polyhedral surface. The data are usually distorted by measuring errors like noise, aliasing, outliers, calibration and registration errors, etc., so that they have to be filtered. Calibration and registration errors first appear after merging of different views. As the merged data are no longer represented on a grid, conventional filters for digital signal processing are not applicable. We introduce a new approach for modeling and smoothing scattered data based on an approximation of a mesh of circular arcs. This new method enables interpolation of curved surfaces using solely the vertex position and the associated vertex normals of a polyhedral mesh. The new smoothing filter is specifically adapted to the requirements of geometric data, as it minimizes curvature variations. In contrast to linear filters, undesired surface undulations are avoided, which is an important pre- condition for NC milling and rendering.
In this paper, the analysis of the measurement errors in an optical sensor for the achievement of 3D range images is presented together with the procedures developed for their compensation. The optical sensor performs the 3D measurement by means of the active stereo vision approach in which the correspondence problem is solved by means of the projection of structured light, basically consisting of patterns of fringes with rectangular profile. Two measurement errors have been identified: the former, thereafter called 'waviness,' is intrinsic to the rectangular profile of the fringes. This error has been dramatically reduced by means of a filtering block, based on a cascade of two Butterworth low-pass IIR filters. The latter, thereafter called 'slope,' is due to the crossed axes optical geometry of the system, which results into a stripe broadening of the projected fringes. The compensation of this error has been achieved by taking into account the orientation of the optical devices of the system and by mapping light directions into measurement coordinates. In this paper, the main features of the procedures developed to minimize these errors are presented and some interesting experimental results are shown.
A solution to the remote three-dimensional (3-D) measurement problem is presented for a dynamic system given a sequence of two-dimensional (2-D) intensity images of a moving object. The 3-D transformation is modeled as a nonlinear stochastic system with the state estimate providing the six-degree-of-freedom motion and position values as well as structure. The stochastic model uses the iterated extended Kalman filter (IEKF) as a nonlinear estimator and a screw representation of the 3-D transformation based on dual quaternions. Dual quaternions, whose elements are dual numbers, provide a means to represent both rotation and translation in a unified notation. Linear object features, represented as dual vectors, are transformed using the dual quaternion transformation and are then projected to linear features in the image plane. The method has been implemented and tested with both simulated and actual experimental data. Simulation results are provided, along with comparisons to a point-based IEKF method using rotation and translation, to show the relative advantages of this method. Experimental results from testing using a camera mounted on the end effector of a robot arm are also given.