Hand-held 3D-scanner for large surface registration

L'objectiu d'aquesta tesi és l'estudi de les diferents tècniques per alinear vistes tridimensionals. Aquest estudi ens ha permès detectar els principals problemes de les tècniques existents, aprotant una solució novedosa i contribuint resolent algunes de les mancances detectades especialment en l'alineament de vistes a temps real. Per tal d'adquirir les esmentades vistes, s'ha dissenyat un sensor 3D manual que ens permet fer adquisicions tridimensionals amb total llibertat de moviments. Així mateix, s'han estudiat les tècniques de minimització global per tal de reduir els efectes de la propagació de l'error.

Contributions to 3D Data Registration and Representation

Nowadays, new computers generation provides a high performance that enables to build computationally expensive computer vision applications applied to mobile robotics. Building a map of the environment is a common task of a robot and is an essential part to allow the robots to move through these environments. Traditionally, mobile robots used a combination of several sensors from different technologies. Lasers, sonars and contact sensors have been typically used in any mobile robotic architecture, however color cameras are an important sensor due to we want the robots to use the same information that humans to sense and move through the different environments. Color cameras are cheap and flexible but a lot of work need to be done to give robots enough visual understanding of the scenes. Computer vision algorithms are computational complex problems but nowadays robots have access to different and powerful architectures that can be used for mobile robotics purposes. The advent of low-cost RGB-D sensors like Microsoft Kinect which provide 3D colored point clouds at high frame rates made the computer vision even more relevant in the mobile robotics field. The combination of visual and 3D data allows the systems to use both computer vision and 3D processing and therefore to be aware of more details of the surrounding environment. The research described in this thesis was motivated by the need of scene mapping. Being aware of the surrounding environment is a key feature in many mobile robotics applications from simple robotic navigation to complex surveillance applications. In addition, the acquisition of a 3D model of the scenes is useful in many areas as video games scene modeling where well-known places are reconstructed and added to game systems or advertising where once you get the 3D model of one room the system can add furniture pieces using augmented reality techniques. In this thesis we perform an experimental study of the state-of-the-art registration methods to find which one fits better to our scene mapping purposes. Different methods are tested and analyzed on different scene distributions of visual and geometry appearance. In addition, this thesis proposes two methods for 3d data compression and representation of 3D maps. Our 3D representation proposal is based on the use of Growing Neural Gas (GNG) method. This Self-Organizing Maps (SOMs) has been successfully used for clustering, pattern recognition and topology representation of various kind of data. Until now, Self-Organizing Maps have been primarily computed offline and their application in 3D data has mainly focused on free noise models without considering time constraints. Self-organising neural models have the ability to provide a good representation of the input space. In particular, the Growing Neural Gas (GNG) is a suitable model because of its flexibility, rapid adaptation and excellent quality of representation. However, this type of learning is time consuming, specially for high-dimensional input data. Since real applications often work under time constraints, it is necessary to adapt the learning process in order to complete it in a predefined time. This thesis proposes a hardware implementation leveraging the computing power of modern GPUs which takes advantage of a new paradigm coined as General-Purpose Computing on Graphics Processing Units (GPGPU). Our proposed geometrical 3D compression method seeks to reduce the 3D information using plane detection as basic structure to compress the data. This is due to our target environments are man-made and therefore there are a lot of points that belong to a plane surface. Our proposed method is able to get good compression results in those man-made scenarios. The detected and compressed planes can be also used in other applications as surface reconstruction or plane-based registration algorithms. Finally, we have also demonstrated the goodness of the GPU technologies getting a high performance implementation of a CAD/CAM common technique called Virtual Digitizing.

3D-2D image registration by nonlinear regression

We propose a 3D-2D image registration method that relates image features of 2D projection images to the transformation parameters of the 3D image by nonlinear regression. The method is compared with a conventional registration method based on iterative optimization. For evaluation, simulated X-ray images (DRRs) were generated from coronary artery tree models derived from 3D CTA scans. Registration of nine vessel trees was performed, and the alignment quality was measured by the mean target registration error (mTRE). The regression approach was shown to be slightly less accurate, but much more robust than the method based on an iterative optimization approach.

Assessment of spline-based 2D-3D registration for image-guided spine surgery

2D-3D registration of pre-operative 3D volumetric data with a series of calibrated and undistorted intra-operative 2D projection images has shown great potential in CT-based surgical navigation because it obviates the invasive procedure of the conventional registration methods. In this study, a recently introduced spline-based multi-resolution 2D-3D image registration algorithm has been adapted together with a novel least-squares normalized pattern intensity (LSNPI) similarity measure for image guided minimally invasive spine surgery. A phantom and a cadaver together with their respective ground truths were specially designed to experimentally assess possible factors that may affect the robustness, accuracy, or efficiency of the registration. Our experiments have shown that it is feasible for the assessed 2D-3D registration algorithm to achieve sub-millimeter accuracy in a realistic setup in less than one minute.

Methodology based on registration techniques for representing subjects and their deformations acquired from general purpose 3D sensors

In this thesis a methodology for representing 3D subjects and their deformations in adverse situations is studied. The study is focused in providing methods based on registration techniques to improve the data in situations where the sensor is working in the limit of its sensitivity. In order to do this, it is proposed two methods to overcome the problems which can difficult the process in these conditions. First a rigid registration based on model registration is presented, where the model of 3D planar markers is used. This model is estimated using a proposed method which improves its quality by taking into account prior knowledge of the marker. To study the deformations, it is proposed a framework to combine multiple spaces in a non-rigid registration technique. This proposal improves the quality of the alignment with a more robust matching process that makes use of all available input data. Moreover, this framework allows the registration of multiple spaces simultaneously providing a more general technique. Concretely, it is instantiated using colour and location in the matching process for 3D location registration.

Medical image registration based on random line sampling

One of the key aspects in 3D-image registration is the computation of the joint intensity histogram. We propose a new approach to compute this histogram using uniformly distributed random lines to sample stochastically the overlapping volume between two 3D-images. The intensity values are captured from the lines at evenly spaced positions, taking an initial random offset different for each line. This method provides us with an accurate, robust and fast mutual information-based registration. The interpolation effects are drastically reduced, due to the stochastic nature of the line generation, and the alignment process is also accelerated. The results obtained show a better performance of the introduced method than the classic computation of the joint histogram

JULIDE: a software tool for 3D reconstruction and statistical analysis of autoradiographic mouse brain sections.

In this article we introduce JULIDE, a software toolkit developed to perform the 3D reconstruction, intensity normalization, volume standardization by 3D image registration and voxel-wise statistical analysis of autoradiographs of mouse brain sections. This software tool has been developed in the open-source ITK software framework and is freely available under a GPL license. The article presents the complete image processing chain from raw data acquisition to 3D statistical group analysis. Results of the group comparison in the context of a study on spatial learning are shown as an illustration of the data that can be obtained with this tool.

A fast 3D approach for coronary MRA.

Two-dimensional (2D)-breath-hold coronary magnetic resonance angiography (MRA) has been shown to be a fast and reliable method to depict the proximal coronary arteries. Recent developments, however, allow for free-breathing navigator gated and navigator corrected three-dimensional (3D) coronary MRA. These 3D approaches have potential for improved signal-to-noise ratio (SNR) and allow for the acquisition of adjacent thin slices without the misregistration problems known from 2D approaches. Still, a major impediment of a 3D acquisition is the increased scan time. The purpose of this study was the implementation of a free-breathing navigator gated and corrected ultra-fast 3D coronary MRA technique, which allows for scan times of less than 5 minutes. Twelve healthy adult subjects were examined in the supine position using a navigator gated and corrected ECG triggered ultra-fast 3D interleaved gradient echo planar imaging sequence (TFE-EPI). A 3D slab, consisting of 20 slices with a reconstructed slice thickness of 1.5 mm, was acquired with free-breathing. The diastolic TFE-EPI acquisition block was preceded by a T2prep pre-pulse, a diaphragmatic navigator pulse, and a fat suppression pre-pulse. With a TR of 19 ms and an effective TE of 5.4 ms, the duration of the data acquisition window duration was 38 ms. The in-plane spatial resolution was 1.0-1.3 mm*1.5-1.9 mm. In all cases, the entire left main (LM) and extensive portions of the left anterior descending (LAD) and right coronary artery (RCA) could be visualized with an average scan time for the entire 3D-volume data set of 2:57 +/- 0:51 minutes. Average contiguous vessel length visualized was 53 +/- 11 mm (range: 42 to 75 mm) for the LAD and 84 +/- 14 mm (range: 62 to 112 mm) for the RCA. Contrast-to-noise between coronary blood and myocardium was 5.0 +/- 2.3 for the LM/LAD and 8.0 +/- 2.9 for the RCA, resulting in an excellent suppression of myocardium. We present a new approach for free-breathing 3D coronary MRA, which allows for scan times superior to corresponding 2D coronary MRA approaches, and which takes advantage of the enhanced SNR of 3D acquisitions and the post-processing benefits of thin adjacent slices. The robust image quality and the short average scanning time suggest that this approach may be useful for screening the major coronary arteries or identification of anomalous coronary arteries. J. Magn. Reson. Imaging 1999;10:821-825.

Medical image registration based on random line sampling

One of the key aspects in 3D-image registration is the computation of the joint intensity histogram. We propose a new approach to compute this histogram using uniformly distributed random lines to sample stochastically the overlapping volume between two 3D-images. The intensity values are captured from the lines at evenly spaced positions, taking an initial random offset different for each line. This method provides us with an accurate, robust and fast mutual information-based registration. The interpolation effects are drastically reduced, due to the stochastic nature of the line generation, and the alignment process is also accelerated. The results obtained show a better performance of the introduced method than the classic computation of the joint histogram

Large-Scale Surface registration

The first part of this work presents an accurate analysis of the most relevant 3D registration techniques, including initial pose estimation, pairwise registration and multiview registration strategies. A new classification has been proposed, based on both the applications and the approach of the methods that have been discussed. The main contribution of this thesis is the proposal of a new 3D multiview registration strategy. The proposed approach detects revisited regions obtaining cycles of views that are used to reduce the inaccuracies that may exist in the final model due to error propagation. The method takes advantage of both global and local information of the registration process, using graph theory techniques in order correlate multiple views and minimize the propagated error by registering the views in an optimal way. The proposed method has been tested using both synthetic and real data, in order to show and study its behavior and demonstrate its reliability.

Single fluoroscopic image based 2D/3D reconstruction of a scaled, patient-specific vertebral model

Purpose Accurate three-dimensional (3D) models of lumbar vertebrae can enable image-based 3D kinematic analysis. The common approach to derive 3D models is by direct segmentation of CT or MRI datasets. However, these have the disadvantages that they are expensive, timeconsuming and/or induce high-radiation doses to the patient. In this study, we present a technique to automatically reconstruct a scaled 3D lumbar vertebral model from a single two-dimensional (2D) lateral fluoroscopic image. Methods Our technique is based on a hybrid 2D/3D deformable registration strategy combining a landmark-to-ray registration with a statistical shape model-based 2D/3D reconstruction scheme. Fig. 1 shows different stages of the reconstruction process. Four cadaveric lumbar spine segments (total twelve lumbar vertebrae) were used to validate the technique. To evaluate the reconstruction accuracy, the surface models reconstructed from the lateral fluoroscopic images were compared to the associated ground truth data derived from a 3D CT-scan reconstruction technique. For each case, a surface-based matching was first used to recover the scale and the rigid transformation between the reconstructed surface model Results Our technique could successfully reconstruct 3D surface models of all twelve vertebrae. After recovering the scale and the rigid transformation between the reconstructed surface models and the ground truth models, the average error of the 2D/3D surface model reconstruction over the twelve lumbar vertebrae was found to be 1.0 mm. The errors of reconstructing surface models of all twelve vertebrae are shown in Fig. 2. It was found that the mean errors of the reconstructed surface models in comparison to their associated ground truths after iterative scaled rigid registrations ranged from 0.7 mm to 1.3 mm and the rootmean squared (RMS) errors ranged from 1.0 mm to 1.7 mm. The average mean reconstruction error was found to be 1.0 mm. Conclusion An accurate, scaled 3D reconstruction of the lumbar vertebra can be obtained from a single lateral fluoroscopic image using a statistical shape model based 2D/3D reconstruction technique. Future work will focus on applying the reconstructed model for 3D kinematic analysis of lumbar vertebrae, an extension of our previously-reported imagebased kinematic analysis. The developed method also has potential applications in surgical planning and navigation.

A study of a soft computing based method for 3D scenario reconstruction

Several recent works deal with 3D data in mobile robotic problems, e.g., mapping. Data comes from any kind of sensor (time of flight, Kinect or 3D lasers) that provide a huge amount of unorganized 3D data. In this paper we detail an efficient approach to build complete 3D models using a soft computing method, the Growing Neural Gas (GNG). As neural models deal easily with noise, imprecision, uncertainty or partial data, GNG provides better results than other approaches. The GNG obtained is then applied to a sequence. We present a comprehensive study on GNG parameters to ensure the best result at the lowest time cost. From this GNG structure, we propose to calculate planar patches and thus obtaining a fast method to compute the movement performed by a mobile robot by means of a 3D models registration algorithm. Final results of 3D mapping are also shown.