Computer vision lacks a solid definition, as it endorses many subjects, among which 3D reconstruction algorithms hold a prominent position. Such algorithms, focusing on the domain of machine vision, advanced quick- ly starting from zero, using linear models as mathematical background. Most of them were developed for speed and real time processing rather than metric accuracy (S ZEliSKi 2011). It soon become apparent that photogrammetry could be benefited by computer vision algo- rithms or vice versa. Nowadays, image based Summary: Recent advances in Structure-from- Motion (SfM) and multi-viewstereo (MVS) tech- niques are getting a lot of attention in the 3D mod- eling community, as they provide a sophisticated, versatile and rapid way to compute dense 3D point clouds. Many applications in archeology and aerial, surface or underwater surveys have been reported based in the combination of SfM and MVS. In ad- dition, they become more popular and widespread as commercial applications and Internet services are emerging. Nevertheless, there is no known way to evaluate the precision of the extracted point cloud. This paper presents a method for estimating the precision of each point in the final point cloud generated by open source MVS, as it is applied to the Parthenon frieze in the Acropolis Museum.
At the moment the reconstruction requires a high-end graphics card to run in real-time. In the project “Light-field camera-based Embedded Sensor Platform” TU Ilmenau will implement the image reconstruction for a Raytrix R8 camera in a Xilinx Zynq Ultrascale+ board. The project partners in parallel develop software applications for ambient assisted living and road condition measurement. Both fields are economically lucrative and would profit from depth information to extend the capabilities of the up to now 2d-based systems. In this paper the focus will lie on the implementation of 3d reconstruction of the scene geometry rather than the applications. The depth calculation is analogous to multi-viewstereo approaches and shares some of their conditions and requirements although the magnitude of their influence is not always identical. In section “Requirements for passive depth estimation” the boundaries of light-field cameras and multi-view-stereo approaches are explained and illustrated.
In this work, we propose a novel approach to prioritize the depth map computation of multi-viewstereo (MVS) to obtain compact 3D point clouds of high quality and completeness at low computational cost. Our prioritization approach operates before the MVS algorithm is executed and consists of two steps. In the first step, we aim to find a good set of matching partners for each view. In the second step, we rank the resulting view clusters (i.e. key views with matching partners) according to their impact on the fulfillment of desired quality parameters such as completeness, ground resolution and accuracy. Additional to geometric analysis, we use a novel machine learning technique for training a confidence predictor. The purpose of this confidence predictor is to estimate the chances of a successful depth reconstruction for each pixel in each image for one specific MVS algorithm based on the RGB images and the image constellation. The underlying machine learning technique does not require any ground truth or manually labeled data for training, but instead adapts ideas from depth map fusion for providing a supervision signal. The trained confidence predictor allows us to evaluate the quality of image constellations and their potential impact to the resulting 3D reconstruction and thus builds a solid foundation for our prioritization approach. In our experiments, we are thus able to reach more than 70% of the maximal reachable quality fulfillment using only 5% of the available images as key views. For evaluating our approach within and across different domains, we use two completely different scenarios, i.e. cultural heritage preservation and reconstruction of single family houses. Keywords: Multi-ViewStereo, Machine Learning, Confidence Measures, View Prioritization, Image Clustering, View Cluster Ranking
In this paper, we address the problem of UAV-based im- age acquisition for dense monocular 3D reconstruction with high-resolution images at close range. The aim is to acquire images in such a way that they are suited for processing with an offline dense multi-viewstereo (MVS) algorithm, while at the same time fulfilling a set of quality require- ments. These requirements include coverage, ground reso- lution and 3D accuracy and can be assessed geometrically. However, determining how well the images are suited for a specific MVS algorithm is much harder to model. To ex- tract depth from 2D images, MVS approaches have to es- tablish correspondences between the images. To solve this challenging task every MVS approach has to make some assumptions. These assumptions vary from approach to ap- proach, but the most popular assumptions include saliency, ∗ The research leading to these results has received funding from the
In a training phase, data of the sought objects are collected by the same sens- ing process that is used later for recognition. In the present case, we collect range data points produced by the stereo algorithm. Depending on the object’s complexity, between two and, say, ten different views of the object are acquired. Different views are registered in an object coordinate system by an external pose measurement device, e.g., a robot. Alternatively, given sufficient overlap between the data sets, a registration of the different views may be achieved by the very same algorithm used later for pose estimation.
Sufficient TR signal was available in 94% of the patients in the RV modified apical four chamber view, and in 77% of the patients in the apical long axis view of RV inflow. Sensitivity for correct diagnosis of PH was 79.6 and 66.7% respectively, AUC for correct diagnosis was 0.85 and 0.83 respectively. The apical long axis view of RV inflow seems to be most important aside from the apical four chamber view. Sensitivity was low in the remaining three imag- ing windows, sufficient TR signal was available in < 50% of the cases in these views. However, if available, these signals correlated significantly with invasively measured PASP and mPAP. Combining all five measurements and using the maximal TR velocity lead to an increase in sen- sitivity to 87% without losing specificity (90.9%). The AUC increased to 0.89. According to these results, the RV modified apical four-chamber view was the single view with best sensitivity and with best availability. Especially in cases with poor signal quality additional views helped make the correct diagnosis.
Die Automatisierung der Synchronisation der Daten wird mittels eines Cronjob realisiert. Dazu wurde eine cron- Tabelle erzeugt, die die Downloadskripte für die einzelnen Instrumente täglich um 20 Uhr startet. Der Download der Daten beginnt täglich um 20 Uhr, weil die Kassen des Planetariums spätestens um 20 Uhr schließen. Wie vorab in Kapitel 3 beschrieben, ist der STEREO Server am Planetarium über die gleiche Leitung mit dem Internet verbunden, wie die Rechner des Kassensystems. Um das Kassensystems und das tägliche Geschäft des Planetariums nicht zu beeinflussen, wurde der Download der Daten auf den späten Abend verschoben. Auf Grund der in Kapitel 3 erwähnten Beschränkung des Downloadlimits werden bei der Spiegelung der Daten die Skripte für die Instrumente mit erhöhtem Datenvolumen aus der Crontab heraus mehrmals gestartet. Vorhandene Daten werden dabei nicht überschrieben bzw. ersetzt, was die Backupzeiten wesentlich verkürzt und sicherstellt, dass der Download bis zur Öffnung des Planetariums abgeschlossen ist. Tabelle 10 gibt einen Überblick über die Downloadskripte und wann sie gestartet werden.
we restrict the maximum number of elements contained by a single grid cell to n max . The reason for this limitation is twofold. First, for high redundant datasets the assignment of each point of the generated depth maps would induce tremendous memory requirements. Second, particularly for data sets captured by wide angle camera systems, a considerable number of points representing 3D structure, as for example fa¸cades points, are extracted. These points hinder the generation of elevation maps since only 2.5D surfaces can be represented and generally the highest elevations in each grid cell are of importance. Not filtering the points representing 3D structure can lead to scattered building edges as visualized in figure 5.4 (left). In the example area marked by the green rectangle facade points are extracted below the roof. This causes median filtered elevations to alter between facade and roof points. Within our filter approach we set n max to the number of average detections per grid cell. This value can be derived from the initialization process. If a candidate point is added to a grid cell which already contains n max elevations, existing elevations and the candidate are sorted and the lowest elevation is discarded. After all points have been assigned to the grid, we employ cell- wise median filtering which defines the final elevation. Figure 5.4 (right) shows the result eliminating the scattering effects by limiting the maximal elements per cell and therefore favoring the highest (roof) points. To guarantee robust results we only consider cells for which at least n min candidates were assigned. Due to the moderate number of outliers produced by the multi-baseline approach n min = 3 is typically sufficient. The final elevation maps are then subject to speckle filtering and median filtering using a kernel size of 3 × 3. The algorithm for the generation of elevation maps is summarized in figure 5.1. Limiting elevation buffers to n max reduces the memory footprint, however, large scale projects require a tiled processing scheme. Thereby points generated by the multi-baseline triangulation are not stored image-wise but tile-wise. Practically points are appended to a file containing all points within a bounding box defining a certain spatial extend. The proposed fusion strategy for elevation maps is then carried out for each single tile. This way we guarantee to load and process each point only once to avoid additional input/output as well as processing overhead. Since the filtering is applied in the direction orthogonal to the surface filtering artifacts at tile borders are prevented.
Die Hörereignisrichtungen auf der Lautsprecherbasis a bis i ergeben sich*) für einen hier angenommenen Ausdeh- nungsbereich eines Klangkörpers von 90° vom Stereo-Mikrofon aus gesehen. Wird das Mikrofonsystem z. B. näher an das Ensemble gestellt, dann wird die "Orchesterausdehnung" größer als 90°. Die Hörereignisrichtungen erscheinen dann weiter nach außen in die Richtung der Lautsprecher gezogen zu sein.
Die Hörereignisrichtungen auf der Lautsprecherbasis a bis i ergeben sich*) für einen hier angenommenen Ausdeh- nungsbereich eines Klangkörpers von 60° vom Stereo-Mikrofon aus gesehen. Wird das Mikrofonsystem z. B. näher an das Ensemble gestellt, dann wird die "Orchesterausdehnung" größer als 60°. Die Hörereignisrichtungen erscheinen dann weiter nach außen in die Richtung der Lautsprecher gezogen zu sein.
In order to carry out the experiments during the 3D-MVPTV development, a former test stand at the Hermann-Rietschel-Institut was modified to become the 3D-MVPTV test stand. As shown in figure 4.1, this test stand has dimensions of 2 .5m × 2.5m × 1.5m. The inlet near to the floor is equipped with three axial fans (Sharkoon System Fan P 120 × 120 × 25mm) in horizontal order, which can supply air for the test stand with a velocity up to circa 0.5m/s. These fans are connected to the power supply through a dimmer unit so that the air velocity flowing into the test stand can be adjusted. The turbulent vortex structures generated by the axial fans are reduced using a flow laminator with honeycomb structures. The outlet opening is placed near the ceiling. In the middle of the room, there is a cylinder with a diameter of 27cm and a height of 56cm. It serves as an optical obstacle which is mainly used to demonstrate the multi-view strength of the 3D-MVPTV. This cylinder can be used at the same time as a heat source, which makes it possible to run experiments with buoyancy-driven flow structures.
Automatic building extraction is one of the most challenging problems in remote sensing image analyses. In addition to 2D information from multi-spectral satellite images, height information from DSMs has received increasingly attention for automatic building extraction. A lot of work has been done by combining use of Airborne LiDAR data and airborne multi-spectral data. However both kinds of datasets are expensive and exhibit low temporal resolution. Therefore, the building extraction from satellite imagery is an interesting research topic.
Die Hörereignisrichtungen auf der Lautsprecherbasis a bis i ergeben sich*) für einen hier angenommenen Ausdeh- nungsbereich eines Klangkörpers von 120° vom Stereo-Mikrofon aus gesehen. Wird das Mikrofonsystem z. B. näher an das Ensemble gestellt, dann wird die "Orchesterausdehnung" größer als 120°. Die Hörereignisrichtungen erscheinen dann weiter nach außen in die Richtung der Lautsprecher gezogen zu sein.
A regenerative RS decodes and re-encodes the received data streams and performs transmit beamforming to transmit the corresponding data streams to the correspond- ing nodes according to the BC strategies. The achievable sum rate for regenerative MGMW relaying is defined by the minimum between the achievable MAC rate at the RS and the achievable BC rate from the RS to the nodes. First, we consider an MMSE-SIC multi-user detector, which is optimum for the uplink MAC, for decoding the data streams of all nodes. Since the RS knows the achievable data rates in the MAC phase prior to BC transmission, we design generalised transmit beamforming for all BC strategies minimising the RS’s transmit power while ensuring that each node receives a particular data stream with a rate equal to the rate received at the RS. Since finding the optimum transmit beamforming minimising the transmit power requires high com- plexity and in some cases the transmit power at the RS is fixed, we design generalised low complexity linear transmit beamforming algorithms for all BC strategies, namely, MF, ZF and MMSE. Also, BCSA transmit beamforming is introduced. It is shown that multicasting-XOR strategy requires least power compared to the other strategies followed by hybrid uni/multicasting, unicasting and multicasting-mSPC. In general, the sum rate performance of regenerative MGMW relaying depends on the chosen BC strategy and the applied transmit beamforming algorithm. Due to its better approach of handling the interference in the network, BCSA transmit beamforming is able to im- prove the performance of regenerative MGMW relaying. In general, multicasting-XOR strategy performs best followed by hybrid uni/multicasting and unicasting strategies. Furthermore, multicasting-XOR outperforms multicasting-mSPC.
In this contribution, the scaled boundary finite element method was used to calculate dispersion curves and mode shapes, as well as to analyse the propagation of guided waves in a plate consisting of an isotropic metal bonded to anisotropic carbon fibre reinforced layered material. The method allows appropriate modes to be identified and their interaction with different damage types to be analysed. The results show that most of the wave energy can be concentrated in a certain layer depending on the mode used, and by that damage present in this respective layer can be detected. The results obtained help to understand the wave propagation in such a multi-layered structure, which may further help guided wave based techniques to be enhanced in the sense of non-destructive testing and structural health monitoring systems to be developed for hybrid composite structures. Moreover, a concept for an SHM system for composite pressure vessels is proposed.
In this paper, we presented a multi-agent freight transport model in which logistics decisions are separated into two different roles: Transport service providers, which create transport chains, and carriers, which plan tours and schedule vehicles. Both agent types can consolidate on their respective level and realise economies of scale. The lowest tier of the model, which contains individual freight vehicles, was integrated into the MATSim traffic simulation to create an integrated model for freight and passenger traffic. Changes in passenger demand, disturbances in the traffic system or policy measures can be picked up by freight drivers and propagated upwards to influence decisions on the levels of vehicle scheduling and transport chain building.
Zusammen mit dem Regisseur plant er weiterhin die möglichen verwendbaren Stereoparameter für jede Einstellung oder Szene. Er ist also auch während der Erstellung eines Storyboards vor Ort, um die möglichen stereoskopischen Ein- stellungen zu besprechen. Es wird ein Tiefenskript erstellt, an dem sich während der Produktion Regie, Kameraabteilung und der Stereograph selber orientieren können. Da zur Zeit viele Filmschaffende noch keine oder wenig Erfahrung mit dem Produzieren von stereoskopischen Bildern haben, ist eine erfahrene Fach- kraft, die bereits während der Planung oder sogar schon während des Verfas- sens des Drehbuchs zur Seite steht, äußerst wichtig. 68 Während des Drehs be- treut der Stereograph den Aufbau der stereoskopischen Kamera und der übrigen, für die stereoskopische Produktion relevanten, Technik. Darunter fallen auch das Wechseln von Objektiven, das bei einem Stereo-3D-Rig wesentlich länger als bei einem 2D-Dreh dauern kann. Dies und die nachträgliche Neueinstellung der rele- vanten Werte geschieht in Zusammenarbeit mit dem Kameraassistenten. Weiter-
[Aleksi´c et al., 2012a] Aleksi´c, J., Alvarez, E. A., Antonelli, L. A., Antoranz, P., Asensio, M., Backes, M., Barrio, J. A., Bastieri, D., Becerra Gonz´alez, J., Bednarek, W., Berdyugin, A., Berger, K., Bernardini, E., Biland, A., Blanch, O., Bock, R. K., Boller, A., Bonnoli, G., Borla Tridon, D., Braun, I., Bretz, T., Ca˜nellas, A., Carmona, E., Carosi, A., Colin, P., Colombo, E., Contreras, J. L., Cortina, J., Cossio, L., Covino, S., Dazzi, F., de Angelis, A., de Caneva, G., de Cea Del Pozo, E., de Lotto, B., Delgado Mendez, C., Diago Ortega, A., Doert, M., Dom´ınguez, A., Dominis Prester, D., Dorner, D., Doro, M., Elsaesser, D., Ferenc, D., Fon- seca, M. V., Font, L., Fruck, C., Garc´ıa L´opez, R. J., Garczarczyk, M., Garrido, D., Giavitto, G., Godinovi´c, N., Hadasch, D., H¨afner, D., Herrero, A., Hildebrand, D., H¨ohne-M¨onch, D., Hose, J., Hrupec, D., Huber, B., Jogler, T., Kellermann, H., Klepser, S., Kr¨ahenb¨uhl, T., Krause, J., La Barbera, A., Lelas, D., Leonardo, E., Lindfors, E., Lombardi, S., L´opez, M., L´opez-Oramas, A., Lorenz, E., Makariev, M., Maneva, G., Mankuzhiyil, N., Mannheim, K., Maraschi, L., Mariotti, M., Mart´ınez, M., Mazin, D., Meucci, M., Miranda, J. M., Mirzoyan, R., Miyamoto, H., Mold´on, J., Moralejo, A., Munar-Adrover, P., Nieto, D., Nilsson, K., Orito, R., Oya, I., Paneque, D., Paoletti, R., Pardo, S., Paredes, J. M., Partini, S., Pasanen, M., Pauss, F., Perez-Torres, M. A., Persic, M., Peruzzo, L., Pilia, M., Pochon, J., Prada, F., Prada Mo- roni, P. G., Prandini, E., Puljak, I., Reichardt, I., Reinthal, R., Rhode, W., Rib´o, M., Rico, J., R¨ugamer, S., Saggion, A., Saito, K., Saito, T. Y., Salvati, M., Satalecka, K., Scalzotto, V., Scapin, V., Schultz, C., Schweizer, T., Shayduk, M., Shore, S. N., Sillanp¨a¨a, A., Sitarek, J., Snidaric, I., Sobczynska, D., Spanier, F., Spiro, S., Stamatescu, V., Stamerra, A., Steinke, B., Storz, J., Strah, N., Suri´c, T., Takalo, L., Takami, H., Tavecchio, F., Temnikov, P., Terzi´c, T., Tescaro, D., Teshima, M., Tibolla, O., Torres, D. F., Treves, A., Uellenbeck, M., Vankov, H., Vogler, P., Wagner, R. M., Weitzel, Q., Zabalza, V., Zandanel, F., and Zanin, R. (2012a). Performance of the MAGIC stereo system obtained with Crab Nebula data. Astroparticle Physics, 35:435–448.