= Industrial image analysis, = Technologies for life science, = Computer graphics, = Medical image analysis

	ABSTRACTS
	Industrial Image Analysis
	Computer Simulation of Fluorescence Loss in Photobleaching Christian Valdemar Hansen / Southern University of Denmark Fluorescence Loss in Photobleaching (FLIP) is a modern microscopy method for visualization of transport processes in living cells. Although FLIP is widespread, an automated reliable analysis of image data is still lacking. The talk will be based on the paper “Computational Modeling of Fluorescence Loss in Photobleaching” by Chr. V. Hansen, H. J. Schroll and D. Wüstner, that presents a well–posed computational model based on spatially resolved diffusion and transport rates. The model is a reaction–diffusion system, discretized by continuous finite elements. The cell geometry is segmented from FLIP images using an active contours algorithm and the PDE model is subsequently solved in real, two–dimensional geometry. Based on this model, FLIP images are simulated and thus molecular transport in living cells is reliably quantified. A framework for the fast set-up of robot assembly solutions Norbert Kruger / SDU - The Maersk Mc-Kinney Moller institute In my talk, I will present work addressing the problem of efficiently setting-up automated assembly processes with robots. The large set-up times for robot assembly processes that are still required today are the reason for the dominance of manual work in production: only 15% of production is automated today. This then often leads to a move of production into countries with low salaries. The use of robots however could make production more cost-efficient and competitive even if high salaries are paid Set-up times of robot systems are dominated by a number of sub-issues: First, often specialized grippers are used for grasping and manipulation that allow for good force control. These grippers often need to be manually designed or refined for particular objects occurring in the assembly process. Second, it is often required to assure that the position and orientation of objects is predetermined with a high degree of precision. This usually requires specific and often rather expensive machinery for precise positioning. Third, robot grasps and trajectories (including appropriate forces) need to be taught in or programmed which often is done through menu-oriented control devices, a quite tedious procedure. Finally, there does not exist yet stable mechanisms for the re-use of action experience. Usually, such experience stays in the brain of the engineer who sets-up the robot solutions. In my talk, I will introduce a framework for the fast set-up of robot assembly solutions. In this framework, we make use of a set of methods which cover computer vision, dynamic simulation, robot control learning, learning by demonstration, dexterous grasping and simulation, one click calibration as well as the re-use of actions stored in an action library. The framework has been developed in particular in the projects IntellAct (2011-2014) and CARMEN (2013-2017). Realtime 3D Model Acquisition of Deformable Objects - A Step Towards Hyper Flexible Automation Sebastian Nesgaard Jensen / DTU compute In my presentation, I will discuss my project and the motivation for starting it. It is a part of a larger initiative called the Manufacturing Academy of Denmark (MADE), an organization which was recently formed by both private and publics partners. Among it s members are Danish Industry, most major danish universities and production companies such as Aalborg Portland, Vestas and Danish Crown. Their goal is to address pressing, contemporary issues with danish production. The setup and development time of automation has been identified as one of the major issues. Automation is very important as it can make Danish industry competitive with other countries where wages for manual labor are much lower. However the huge initial investments required for introducing automated solution deters many. And when a solution has been obtained it can typically only perform a very narrow range of operations. This makes it suitable only for the task which it was initially developed for. As such when one wants to introduce new automation an entirely new solution has to be obtained with a similar investment. In order to resolve this issue, Working Package 8 (WP8) was formed within MADE. Our goal is to develop automation technology that is very flexible while maintaining the efficiency of contemporary solutions. Meaning that a solution only has to be developed once and can automate many tasks. My contribution will be to develop real-time 3D scanning technology aimed at guiding robotics. During my talk, I will discuss this goal as well as my primary case study which is automation of the packaging of meat products at the Danish Crown facilities. Electrical Simulation of Cochlear Implants based on Patient Specific Shape Modelling Mario Ceresa / Universitat Pompeu Fabra, Barcelona Cochlear implants (CI) proved to be very successful at hearing restoration. However, there is a lack of pre-operative measures that predict the outcomes after implantation. We argue that highly detailed computational models that are specifically tailored for a patient can provide useful information to improve the precision of the nervous system electrode interface In this talk we'll present a procedure to enhance the clinical CT scan taken in the PreOp scenario with a highly detailed computational model. We use a morphological model based on high-definition mCT scans to capture the anatomical variability present in the normal population. A Finite Element problem is defined into this model and then transferred to a real clinical case to augment it and to advice on the output of the surgery. Measuring and Simulating Head-Related Transfer Functions using 3D Scanning and printing Søren Laugesen / Oticon Research Centre All heads and ears are different, but nevertheless most audio devices ? including hearing aids ? are manufactured according to average acoustical data. There are, however, considerable performance improvements to be obtained if the devices were acoustically fitted to the individual. An important part of the individual acoustics is captured in the so-called head-related transfer functions (HRTFs). Direct measurement of HRTFs is possible in the laboratory, but cumbersome in development and completely intractable in clinical practice. Therefore alternative acquisition methods are attractive. Here, the potential of using individual 3D scans is investigated. The scans can either be used to create numerical simulation models to entirely bypass acoustical measurements, or the scans can be used to print 3D models for measurement purposes. The latter approach is particularly useful for children, who find it difficult to sit still for the lengthy HRTF measurements. This talk will address three fundamental questions regarding the viability of using 3D-scans for HRTF acquisition: 1) how accurately can HRTFs be measured under the best possible circumstances? 2) How close to the measured result is an equivalent numerical simulation (using the Finite-Element Method)? 3 How close to the real-human measured result is a measurement taken on a printed model? Improving tomographic reconstruction Emil Bøje Lind Pedersen / DTU Energy In material science, tomography is a useful tool to test and understand hypotheses regarding structural composition. As the engineering efforts approach the nanometer scale, tomography is also pushed to reach higher and higher resolution. On the experimental side, new contrast techniques are developed and high precision is pursued using sample stages with interferometry setups and cryogenic conditions. The experimental efforts should be matched by developments in algorithms and models, in order to extract the maximum information from the tomographic data. I will present a tomographic alignment model, that aim at reaching higher resolution by reducing projection misalignment without fiducial markers and a segmentation model that includes partial volumes effect to access sub voxel density information.
	Computer Graphics
	Topology Optimization for the Masses Morten Nobel-Jørgensen / DTU Compute Topology Optimization is a subfield of structural optimization and deals with optimization of solid (structural) and fluid mechanics. The optimization can improve the quality of a product in terms of improved stiffness, reduced weight and reduced material consumption. Traditional topology optimization software is used as separate step in computer-aided design (CAD) and is targeted a narrow audience of expert users. Through a series of studies, we have investigated how topology optimization can be made more accessible to a general audience with little knowledge of mechanical engineering. Our prototypes focuses both on facilitating learning as well as discovering new ways of using topology optimization in CAD-tools. This improved user interaction gives industrial designers the ability to create better products in a faster and more convenient way. Layered Reconstruction for Defocus and Motion Blur Jacob Munkberg / Intel Corporation, Advanced Rendering Technology Group Light field reconstruction algorithms can substantially decrease the noise in stochastically rendered images. Recent algorithms for defocus blur alone are both fast and accurate. However, motion blur is a considerably more complex type of camera effect, and as a consequence, current algorithms are either slow or too imprecise to use in high quality rendering. We extend previous work on real-time light field reconstruction for defocus blur to handle the case of simultaneous defocus and motion blur. By carefully introducing a few approximations, we derive a very efficient sheared reconstruction filter, which produces high quality images even for a low number of input samples. Our algorithm is temporally robust, and is about two orders of magnitude faster than previous work, making it suitable for both real-time rendering and as a post-processing pass for offline rendering.
	Technologies for Life Science
	Studies on the Influence of Weather Patterns on Regional Pollution Using Statistical Classification Methods Prof. Ahmet Palazoglu / University of California, Davis - Chemical Engineering and Material Science Ozone is a secondary pollutant that can accumulate to unhealthy levels in urban areas given sufficient precursor concentrations and favorable meteorological conditions. Furthermore, fine particulate matter (particles with aerodynamic diameter 2.5 mm or less, denoted PM2.5) pollution has been empirically demonstrated as deleterious to human health. It is important to understand the dynamics of ozone/PM episodes as a function of weather events in order to draw sound policies and abatement strategies. While Air Quality Models (AQMs) coupled with meteorological models provide simulation evidence, there is an abundance of regional pollution and weather data that can be mined to determine patterns and trends that can complement such rigorous studies and provide valuable insight. Such statistical models can also provide forecasts to warn urban populations and to plan for modification of, e.g., driving patterns in a given day. Our studies focused on identifying wind field patterns that are associated with poor (or favorable) pollution characteristics. Using years-long data of hourly wind direction and wind speed measurements in several monitoring stations across northern and central California, we were able to determine key patterns that drive high-pollution levels. Our studies have also extended to the exploration of decadal global patterns (e.g., El Nino) in terms of their influence on pollution events across the US. This talk will introduce some of the methods used for such studies (e.g. PCA, clustering and others) and illustrate them with specific examples. Real Time insights on development in the European labor market Peter Berg Jørgensen / DAMVAD Group By using big data analytics, the specialized policy and economics consultancy DAMVAD, are building a complete database of European Job adds. The DAMVAD Job Ad Database gives a real time perspective on what type of jobs and individual qualifications/competencies, companies is the EU are looking for. It also provides a solid foundation for forecasting labor market development. The project poses many analytical challenges for how to extract valuable information from a large corpus of structured and unstructured data in several languages. In this talk, we will share how we by using statistic matching and machine learning methodology, are able to provide real time insights into the European labor market. Heatmap based tumor heterogeneity assessment in malignant melanoma Anne-Sofie Wessel Lindberg / DTU Compute and Visiopharm Malignant melanoma is one of the most common cancer types. Today, diagnosis is determined using a histologic tissue sample taken from the middle of the tumor. But is it appropriate to assume that this represent the extreme heterogeneous composition of cancer cells? During the presentation we investigate this assumption and develop new methods for describing the heterogeneity through the physical tumor.