Methodologies for visual correspondence

Visual correspondence is a key computer vision task that aims at identifying projections of the same 3D point into images taken either from different viewpoints or at different time instances. This task has been the subject of intense research activities in the last years in scenarios such as object recognition, motion detection, stereo vision, pattern matching, image registration. The approaches proposed in literature typically aim at improving the state of the art by increasing the reliability, the accuracy or the computational efficiency of visual correspondence algorithms. The research work carried out during the Ph.D. course and presented in this dissertation deals with three specific visual correspondence problems: fast pattern matching, stereo correspondence and robust image matching. The dissertation presents original contributions to the theory of visual correspondence, as well as applications dealing with 3D reconstruction and multi-view video surveillance.

Evaluation of surface defect detection in reinforced concrete bridge decks using terrestrial LiDAR

Routine bridge inspections require labor intensive and highly subjective visual interpretation to determine bridge deck surface condition. Light Detection and Ranging (LiDAR) a relatively new class of survey instrument has become a popular and increasingly used technology for providing as-built and inventory data in civil applications. While an increasing number of private and governmental agencies possess terrestrial and mobile LiDAR systems, an understanding of the technology’s capabilities and potential applications continues to evolve. LiDAR is a line-of-sight instrument and as such, care must be taken when establishing scan locations and resolution to allow the capture of data at an adequate resolution for defining features that contribute to the analysis of bridge deck surface condition. Information such as the location, area, and volume of spalling on deck surfaces, undersides, and support columns can be derived from properly collected LiDAR point clouds. The LiDAR point clouds contain information that can provide quantitative surface condition information, resulting in more accurate structural health monitoring. LiDAR scans were collected at three study bridges, each of which displayed a varying degree of degradation. A variety of commercially available analysis tools and an independently developed algorithm written in ArcGIS Python (ArcPy) were used to locate and quantify surface defects such as location, volume, and area of spalls. The results were visual and numerically displayed in a user-friendly web-based decision support tool integrating prior bridge condition metrics for comparison. LiDAR data processing procedures along with strengths and limitations of point clouds for defining features useful for assessing bridge deck condition are discussed. Point cloud density and incidence angle are two attributes that must be managed carefully to ensure data collected are of high quality and useful for bridge condition evaluation. When collected properly to ensure effective evaluation of bridge surface condition, LiDAR data can be analyzed to provide a useful data set from which to derive bridge deck condition information.

In vitro accuracy evaluation of image-guided robot system for direct cochlear access

HYPOTHESIS A previously developed image-guided robot system can safely drill a tunnel from the lateral mastoid surface, through the facial recess, to the middle ear, as a viable alternative to conventional mastoidectomy for cochlear electrode insertion. BACKGROUND Direct cochlear access (DCA) provides a minimally invasive tunnel from the lateral surface of the mastoid through the facial recess to the middle ear for cochlear electrode insertion. A safe and effective tunnel drilled through the narrow facial recess requires a highly accurate image-guided surgical system. Previous attempts have relied on patient-specific templates and robotic systems to guide drilling tools. In this study, we report on improvements made to an image-guided surgical robot system developed specifically for this purpose and the resulting accuracy achieved in vitro. MATERIALS AND METHODS The proposed image-guided robotic DCA procedure was carried out bilaterally on 4 whole head cadaver specimens. Specimens were implanted with titanium fiducial markers and imaged with cone-beam CT. A preoperative plan was created using a custom software package wherein relevant anatomical structures of the facial recess were segmented, and a drill trajectory targeting the round window was defined. Patient-to-image registration was performed with the custom robot system to reference the preoperative plan, and the DCA tunnel was drilled in 3 stages with progressively longer drill bits. The position of the drilled tunnel was defined as a line fitted to a point cloud of the segmented tunnel using principle component analysis (PCA function in MatLab). The accuracy of the DCA was then assessed by coregistering preoperative and postoperative image data and measuring the deviation of the drilled tunnel from the plan. The final step of electrode insertion was also performed through the DCA tunnel after manual removal of the promontory through the external auditory canal. RESULTS Drilling error was defined as the lateral deviation of the tool in the plane perpendicular to the drill axis (excluding depth error). Errors of 0.08 ± 0.05 mm and 0.15 ± 0.08 mm were measured on the lateral mastoid surface and at the target on the round window, respectively (n =8). Full electrode insertion was possible for 7 cases. In 1 case, the electrode was partially inserted with 1 contact pair external to the cochlea. CONCLUSION The purpose-built robot system was able to perform a safe and reliable DCA for cochlear implantation. The workflow implemented in this study mimics the envisioned clinical procedure showing the feasibility of future clinical implementation.

Registration And Visualisation Of Deformation Maps From Terrestrial Radar Interferometry Using Photogrammetry And Structure From Motion

This paper describes a general workflow for the registration of terrestrial radar interferometric data with 3D point clouds derived from terrestrial photogrammetry and structure from motion. After the determination of intrinsic and extrinsic orientation parameters, data obtained by terrestrial radar interferometry were projected on point clouds and then on the initial photographs. Visualisation of slope deformation measurements on photographs provides an easily understandable and distributable information product, especially of inaccessible target areas such as steep rock walls or in rockfall run-out zones. The suitability and error propagation of the referencing steps and final visualisation of four approaches are compared: (a) the classic approach using a metric camera and stereo-image photogrammetry; (b) images acquired with a metric camera, automatically processed using structure from motion; (c) images acquired with a digital compact camera, processed with structure from motion; and (d) a markerless approach, using images acquired with a digital compact camera using structure from motion without artificial ground control points. The usability of the completely markerless approach for the visualisation of high-resolution radar interferometry assists the production of visualisation products for interpretation.

Deep Points Consolidation

In this paper, we present a consolidation method that is based on a new representation of 3D point sets. The key idea is to augment each surface point into a deep point by associating it with an inner point that resides on the meso-skeleton, which consists of a mixture of skeletal curves and sheets. The deep points representation is a result of a joint optimization applied to both ends of the deep points. The optimization objective is to fairly distribute the end points across the surface and the meso-skeleton, such that the deep point orientations agree with the surface normals. The optimization converges where the inner points form a coherent meso-skeleton, and the surface points are consolidated with the missing regions completed. The strength of this new representation stems from the fact that it is comprised of both local and non-local geometric information. We demonstrate the advantages of the deep points consolidation technique by employing it to consolidate and complete noisy point-sampled geometry with large missing parts.

Creating Useful Products From Connecticut's 2000 LIDAR Data Set JHR 08-314 Project 07-2

The State of Connecticut owns a LIght Detection and Ranging (LIDAR) data set that was collected in 2000 as part of the State’s periodic aerial reconnaissance missions. Although collected eight years ago, these data are just now becoming ready to be made available to the public. These data constitute a massive “point cloud”, being a long list of east-north-up triplets in the State Plane Coordinate System Zone 0600 (SPCS83 0600), orthometric heights (NAVD 88) in US Survey feet. Unfortunately, point clouds have no structure or organization, and consequently they are not as useful as Triangulated Irregular Networks (TINs), digital elevation models (DEMs), contour maps, slope and aspect layers, curvature layers, among others. The goal of this project was to provide the computational infrastructure to create a first cut of these products and to serve them to the public via the World Wide Web. The products are available at http://clear.uconn.edu/data/ct_lidar/index.htm.

GPS raw data (control points and ground control points) from the Liguria Region, Borghetto Santo Spirito, Italy

Monitoring the impact of sea storms on coastal areas is fundamental to study beach evolution and the vulnerability of low-lying coasts to erosion and flooding. Modelling wave runup on a beach is possible, but it requires accurate topographic data and model tuning, that can be done comparing observed and modeled runup. In this study we collected aerial photos using an Unmanned Aerial Vehicle after two different swells on the same study area. We merged the point cloud obtained with photogrammetry with multibeam data, in order to obtain a complete beach topography. Then, on each set of rectified and georeferenced UAV orthophotos, we identified the maximum wave runup for both events recognizing the wet area left by the waves. We then used our topography and numerical models to simulate the wave runup and compare the model results to observed values during the two events. Our results highlight the potential of the methodology presented, which integrates UAV platforms, photogrammetry and Geographic Information Systems to provide faster and cheaper information on beach topography and geomorphology compared with traditional techniques without losing in accuracy. We use the results obtained from this technique as a topographic base for a model that calculates runup for the two swells. The observed and modeled runups are consistent, and open new directions for future research.

Biokerosene from Babassu and Camelina Oils: production and properties of their blends with fossil kerosene

Babassu and camelina oils have been transesterified with methanol by the classical homogeneous basic catalysis method with good yields. The babassu fatty acid methyl ester (FAME) has been subjected to fractional distillation at vacuum, and the low boiling point fraction has been blended with two types of fossil kerosene, a straight-run atmospheric distillation cut (hydrotreated) and a commercial Jet-A1. The camelina FAME has been blended with the fossil kerosene without previous distillation. The blends of babassu biokerosene and Jet-A1 have met some of the specifications selected for study of the ASTM D1655 standard: smoke point, density, flash point, cloud point, kinematic viscosity, oxidative stability and lower heating value. On the other hand, the blends of babassu biokerosene and atmospheric distillation cut only have met the density parameter and the oxidative stability. The blends of camelina FAME and atmospheric distillation cut have met the following specifications: density, kinematic viscosity at −20 °C, and lower heating value. With these preliminary results, it can be concluded that it would be feasible to blend babassu and camelina biokerosenes prepared in this way with commercial Jet-A1 up to 10 vol % of the former, if these blends prove to accomplish all the ASTM D1655-09 standards.

Detection and Tracking of Dynamic Objects. A MultiRobot approach to Critical Infrastructures Surveillance.

Despite that Critical Infrastructures (CIs) security and surveillance are a growing concern for many countries and companies, Multi Robot Systems (MRSs) have not been yet broadly used in this type of facilities. This dissertation presents a novel study of the challenges arisen by the implementation of this type of systems and proposes solutions to specific problems. First, a comprehensive analysis of different types of CIs has been carried out, emphasizing the influence of the different characteristics of the facilities in the design of a security and surveillance MRS. One of the most important needs for the surveillance of a CI is the detection of intruders. From a technical point of view this problem can be abstracted as equivalent to the Detection and Tracking of Mobile Objects (DATMO). This dissertation proposes algorithms to solve this specific problem in a CI environment. Using 3D range images of the environment as input data, two detection algorithms for ground robots have been developed. These detection algorithms provide a list of moving objects in the robot detection area. Direct image differentiation and computer vision techniques are used when the robot is static. Alternatively, multi-layer ground reconstructions are compared to detect the dynamic objects when the robot is moving. Since CIs usually spread over large areas, it is very useful to incorporate aerial vehicles in the surveillance MRS. Therefore, a moving object detection algorithm for aerial vehicles has been also developed. This algorithm compares the real optical flow obtained from a down-face oriented camera with an artificial optical flow computed using a RANSAC based homography matrix. Two tracking algorithms have been developed to follow the moving objects trajectories. These algorithms can efficiently handle occlusions and crossings, as well as exchange information among robots. The multirobot tracking can be applied to any type of communication structure: centralized, decentralized or a combination of both. Even more, the developed tracking algorithms are independent of the detection algorithms and could be potentially used with other detection procedures or even with static sensors, such as cameras. In addition, using the 3D point clouds available to the robots, a relative localization algorithm has been developed to improve the position estimation of a given robot with observations from other robots. All the developed algorithms have been extensively tested in different simulated CIs using the Webots robotics simulator. Furthermore, the algorithms have also been validated with real robots operating in real scenarios. In conclusion, this dissertation presents a multirobot approach to Critical Infrastructure Surveillance, mainly focusing on Detecting and Tracking Dynamic Objects.

Sea Bottom Classification by Means of Bathymetric LIDAR Data

Light Detection and Ranging (LIDAR) provides high horizontal and vertical resolution of spatial data located in point cloud images, and is increasingly being used in a number of applications and disciplines, which have concentrated on the exploit and manipulation of the data using mainly its three dimensional nature. Bathymetric LIDAR systems and data are mainly focused to map depths in shallow and clear waters with a high degree of accuracy. Additionally, the backscattering produced by the different materials distributed over the bottom surface causes that the returned intensity signal contains important information about the reflection properties of these materials. Processing conveniently these values using a Simplified Radiative Transfer Model, allows the identification of different sea bottom types. This paper presents an original method for the classification of sea bottom by means of information processing extracted from the images generated through LIDAR data. The results are validated using a vector database containing benthic information derived by marine surveys.

A computational model of the anterior intraparietal area

The monkey anterior intraparietal area (AIP) encodes visual information about three-dimensional object shape that is used to shape the hand for grasping. In robotics a similar role has been played by modules that fit point cloud data to the superquadric family of shapes and its various extensions. We developed a model of shape tuning in AIP based on cosine tuning to superquadric parameters. However, the model did not fit the data well, and we also found that it was difficult to accurately reproduce these parameters using neural networks with the appropriate inputs (modelled on the caudal intraparietal area, CIP). The latter difficulty was related to the fact that there are large discontinuities in the superquadric parameters between very similar shapes. To address these limitations we adopted an alternative shape parameterization based on an Isomap nonlinear dimension reduction. The Isomap was built using gradients and curvatures of object surface depth. This alternative parameterization was low-dimensional (like superquadrics), but data-driven (similar to an alternative clustering approach that is also sometimes used in robotics) and lacked large discontinuities. Isomaps with 16 or more dimensions reproduced the AIP data fairly well. Moreover, we found that the Isomap parameters could be approximated from CIP-like input much more accurately than the superquadric parameters. We conclude that Isomaps, or perhaps alternative dimension reductions of CIP signals, provide a promising model of AIP tuning. We have now started to integrate our model with a robot hand, to explore the efficacy of Isomap shape reductions in grasp planning. Future work will consider dynamics of spike responses and integration with related visual and motor area models.

Auscultación de presas de hormigón con láser escáner terrestre y modelización de las deformaciones del paramento

Se está produciendo en la geodesia un cambio de paradigma en la concepción de los modelos digitales del terreno, pasando de diseñar el modelo con el menor número de puntos posibles a hacerlo con cientos de miles o millones de puntos. Este cambio ha sido consecuencia de la introducción de nuevas tecnologías como el escáner láser, la interferometría radar y el tratamiento de imágenes. La rápida aceptación de estas nuevas tecnologías se debe principalmente a la gran velocidad en la toma de datos, a la accesibilidad por no precisar de prisma y al alto grado de detalle de los modelos. Los métodos topográficos clásicos se basan en medidas discretas de puntos que considerados en su conjunto forman un modelo; su precisión se deriva de la precisión en la toma singular de estos puntos. La tecnología láser escáner terrestre (TLS) supone una aproximación diferente para la generación del modelo del objeto observado. Las nubes de puntos, producto del escaneo con TLS, pasan a ser tratadas en su conjunto mediante análisis de áreas, de forma que ahora el modelo final no es el resultado de una agregación de puntos sino la de la mejor superficie que se adapta a las nubes de puntos. Al comparar precisiones en la captura de puntos singulares realizados con métodos taquimétricos y equipos TLS la inferioridad de estos últimos es clara; sin embargo es en el tratamiento de las nubes de puntos, con los métodos de análisis basados en áreas, se han obtenido precisiones aceptables y se ha podido considerar plenamente la incorporación de esta tecnología en estudios de deformaciones y movimientos de estructuras. Entre las aplicaciones del TLS destacan las de registro del patrimonio, registro de las fases en la construcción de plantas industriales y estructuras, atestados de accidentes y monitorización de movimientos del terreno y deformaciones de estructuras. En la auscultación de presas, comparado con la monitorización de puntos concretos dentro, en coronación o en el paramento de la presa, disponer de un modelo continuo del paramento aguas abajo de la presa abre la posibilidad de introducir los métodos de análisis de deformaciones de superficies y la creación de modelos de comportamiento que mejoren la comprensión y previsión de sus movimientos. No obstante, la aplicación de la tecnología TLS en la auscultación de presas debe considerarse como un método complementario a los existentes. Mientras que los péndulos y la reciente técnica basada en el sistema de posicionamiento global diferencial (DGPS) dan una información continua de los movimientos de determinados puntos de la presa, el TLS permite ver la evolución estacional y detectar posibles zonas problemáticas en todo el paramento. En este trabajo se analizan las características de la tecnología TLS y los parámetros que intervienen en la precisión final de los escaneos. Se constata la necesidad de utilizar equipos basados en la medida directa del tiempo de vuelo, también llamados pulsados, para distancias entre 100 m y 300 m Se estudia la aplicación del TLS a la modelización de estructuras y paramentos verticales. Se analizan los factores que influyen en la precisión final, como el registro de nubes, tipo de dianas y el efecto conjunto del ángulo y la distancia de escaneo. Finalmente, se hace una comparación de los movimientos dados por los péndulos directos de una presa con los obtenidos del análisis de las nubes de puntos correspondientes a varias campañas de escaneos de la misma presa. Se propone y valida el empleo de gráficos patrón para relacionar las variables precisión o exactitud con los factores distancia y ángulo de escaneo en el diseño de trabajos de campo. Se expone su aplicación en la preparación del trabajo de campo para la realización de una campaña de escaneos dirigida al control de movimientos de una presa y se realizan recomendaciones para la aplicación de la técnica TLS a grandes estructuras. Se ha elaborado el gráfico patrón de un equipo TLS concreto de alcance medio. Para ello se hicieron dos ensayos de campo en condiciones reales de trabajo, realizando escaneos en todo el rango de distancias y ángulos de escaneo del equipo. Se analizan dos métodos para obtener la precisión en la modelización de paramentos y la detección de movimientos de estos: el método del “plano de mejor ajuste” y el método de la “deformación simulada”. Por último, se presentan los resultados de la comparación de los movimientos estacionales de una presa arco-gravedad entre los registrados con los péndulos directos y los obtenidos a partir de los escaneos realizados con un TLS. Los resultados muestran diferencias de milímetros, siendo el mejor de ellos del orden de un milímetro. Se explica la metodología utilizada y se hacen consideraciones respecto a la densidad de puntos de las nubes y al tamaño de las mallas de triángulos. A shift of paradigm in the conception of the survey digital models is taking place in geodesy, moving from designing a model with the fewer possible number of points to models of hundreds of thousand or million points. This change has happened because of the introduction of new technologies like the laser scanner, the interferometry radar and the processing of images. The fast acceptance of these new technologies has been due mainly to the great speed getting the data, to the accessibility as reflectorless technique, and to the high degree of detail of the models. Classic survey methods are based on discreet measures of points that, considered them as a whole, form a model; the precision of the model is then derived from the precision measuring the single points. The terrestrial laser scanner (TLS) technology supposes a different approach to the model generation of the observed object. Point cloud, the result of a TLS scan, must be treated as a whole, by means of area-based analysis; so, the final model is not an aggregation of points but the one resulting from the best surface that fits with the point cloud. Comparing precisions between the one resulting from the capture of singular points made with tachometric measurement methods and with TLS equipment, the inferiority of this last one is clear; but it is in the treatment of the point clouds, using area-based analysis methods, when acceptable precisions have been obtained and it has been possible to consider the incorporation of this technology for monitoring structures deformations. Among TLS applications it have to be emphasized those of registry of the cultural heritage, stages registry during construction of industrial plants and structures, police statement of accidents and monitorization of land movements and structures deformations. Compared with the classical dam monitoring, approach based on the registry of a set of points, the fact having a continuous model of the downstream face allows the possibility of introducing deformation analysis methods and behavior models that would improve the understanding and forecast of dam movements. However, the application of TLS technology for dam monitoring must be considered like a complementary method with the existing ones. Pendulums and recently the differential global positioning system (DGPS) give a continuous information of the movements of certain points of the dam, whereas TLS allows following its seasonal evolution and to detect damaged zones of the dam. A review of the TLS technology characteristics and the factors affecting the final precision of the scanning data is done. It is stated the need of selecting TLS based on the direct time of flight method, also called pulsed, for scanning distances between 100m and 300m. Modelling of structures and vertical walls is studied. Factors that influence in the final precision, like the registry of point clouds, target types, and the combined effect of scanning distance and angle of incidence are analyzed. Finally, a comparison among the movements given by the direct pendulums of a dam and the ones obtained from the analysis of point clouds is done. A new approach to obtain a complete map-type plot of the precisions of TLS equipment based on the direct measurement of time of flight method at midrange distances is presented. Test were developed in field-like conditions, similar to dam monitoring and other civil engineering works. Taking advantage of graphic semiological techniques, a “distance - angle of incidence” map based was designed and evaluated for field-like conditions. A map-type plot was designed combining isolines with sized and grey scale points, proportional to the precision values they represent. Precisions under different field conditions were compared with specifications. For this purpose, point clouds were evaluated under two approaches: the standar "plane-of-best-fit" and the proposed "simulated deformation”, that showed improved performance. These results lead to a discussion and recommendations about optimal TLS operation in civil engineering works. Finally, results of the comparison of seasonal movements of an arc-gravity dam between the registered by the direct pendulums ant the obtained from the TLS scans, are shown. The results show differences of millimeters, being the best around one millimeter. The used methodology is explained and considerations with respect to the point cloud density and to the size of triangular meshes are done.

Three-dimensional planar model estimation using multi-constraint knowledge based on k-means and RANSAC

Plane model extraction from three-dimensional point clouds is a necessary step in many different applications such as planar object reconstruction, indoor mapping and indoor localization. Different RANdom SAmple Consensus (RANSAC)-based methods have been proposed for this purpose in recent years. In this study, we propose a novel method-based on RANSAC called Multiplane Model Estimation, which can estimate multiple plane models simultaneously from a noisy point cloud using the knowledge extracted from a scene (or an object) in order to reconstruct it accurately. This method comprises two steps: first, it clusters the data into planar faces that preserve some constraints defined by knowledge related to the object (e.g., the angles between faces); and second, the models of the planes are estimated based on these data using a novel multi-constraint RANSAC. We performed experiments in the clustering and RANSAC stages, which showed that the proposed method performed better than state-of-the-art methods.

SMR analysis using 3DPC datasets

Datasets and results of the paper: Characterization of rock slopes through slope mass rating using 3D point clouds, Riquelme et al 2016, IJRMMS.

Automatic Identification of Large Fragments in a Pile of Broken Rock Using a Time-of-Flight Camera

This paper presents a solution to part of the problem of making robotic or semi-robotic digging equipment less dependant on human supervision. A method is described for identifying rocks of a certain size that may affect digging efficiency or require special handling. The process involves three main steps. First, by using range and intensity data from a time-of-flight (TOF) camera, a feature descriptor is used to rank points and separate regions surrounding high scoring points. This allows a wide range of rocks to be recognized because features can represent a whole or just part of a rock. Second, these points are filtered to extract only points thought to belong to the large object. Finally, a check is carried out to verify that the resultant point cloud actually represents a rock. Results are presented from field testing on piles of fragmented rock. Note to Practitioners—This paper presents an algorithm to identify large boulders in a pile of broken rock as a step towards an autonomous mining dig planner. In mining, piles of broken rock can contain large fragments that may need to be specially handled. To assess rock piles for excavation, we make use of a TOF camera that does not rely on external lighting to generate a point cloud of the rock pile. We then segment large boulders from its surface by using a novel feature descriptor and distinguish between real and false boulder candidates. Preliminary field experiments show promising results with the algorithm performing nearly as well as human test subjects.