Search

This paper presents a comparative assessment of two-dimensional convolutional neural networks (2DCNN) and boosting methods for regression-based textile whiteness estimation, applied to high resolution images of textiles of an industrial cotton textiles producer, labeled with whiteness values, thus enabling supervised learning. The images were taken under various lighting conditions. Concerning the machine learning methods, Random Forest and XGBoost were the selected and tested boosting techniques on which model hyper-parameter tuning was applied, whereas regarding the 2DCNN architectures, the known from literature ColorNet architecture was selected and a more shallow one, called WERegNet, was introduced. Data augmentation was applied during pre-processing, due to the limited amount of available data. Based on the simulation results, the WERegNet architecture surpasses ColorNet and XGBoost in terms of performance, while it is comparable with Random Forest on test set, based on model selection measure Normalized Root Mean Squared Error.

This work presents a cognitive analytics platform for anomaly detection which is capable to handle, analyze and exploit resourcefully machine data from a shop-floor of factory, so as to support the emerging and growing needs of manufacturing industry. The introduced system contributes to industrial digitization and creation of smart factories by providing a generic platform which is a complete solution supporting standards-based factory connectivity, data management, various AI models training and comparisons, live predictions and real-time visualizations. The proposed system is built on a micro-service architecture, in order to be extendable and adaptive over time, and contains three core modules, the Data Acquisition, the Knowledge Management and the Predictive maintenance, which contribute to machine failure prediction and activate predictive maintenance procedures, to efficient production schemes and decision making, to monitor anomalies and handle unforeseen conditions, to predict future behaviours on time series etc. The proposed platform utilizes continuous re-training mechanisms enabling a self learning approach for the delivery of AI solutions, usable also for various production data, guaranteeing the quality of results without continuous monitoring and human-resources allocation for AI models’ retraining. This cognitive platform is supported by machine learning techniques and deep learning architectures in order to achieve the desired performance in the management of factory processes and needs. All the information generated by the proposed platform is provided to the end user through a user interface that utilizes advanced visualization techniques.

Energy sector stakeholders, such as Distribution System Operators (DSO) or Aggregators take advantage of improved forecasting methods. Increased forecasting accuracy facilitates handling energy imbalances between generation and consumption. It also supports Smart Grid framework processes, such as Demand-Side or Demand Response Management (DRM). This paper presents a novel approach for One-Step-Ahead Energy Load Forecasting (OSA-ELF), considering several techniques. It utilizes historical data from a state-of-the-art nearly Zero Energy Building (nZEB) smart home, performing multiple tests for improved ELF. It focuses on OSA aspects of ELF, yet it can be utilized regardless of the time resolution. It predicts the “next step” value, regardless of the step’s duration (15-minutes, one-hour, one-day etc.) with high accuracy, and can be used for a wide variety of forecasting applications. To that end, fine-tuned ensemble methods and forecasting algorithms were utilized for experimenting with short term ELF. Forecasting evaluation produced good results with regards to popular accuracy (MAPE, SMAPE and RMSE) and an Execution Time (ET) metrics.

There is a quite intensive research for Cloud simulators in the recent years, mainly due to the fact that the need for powerful computational resources has led organizations to use cloud resources instead of acquiring and maintaining private servers. In order to test and optimize the strategies that are being used on cloud resources, cloud simulators have been developed since the simulation cost is substantially smaller than experimenting on real cloud environments. Several cloud simulation frameworks have been proposed during the last years, focusing on various components of the cloud resources. In this paper, a survey on cloud simulators is conducted, in order to examine the different models that have been used for the hardware components that constitute a cloud data center. Focus is given on the energy models that have been proposed for the prediction of the energy consumption of data center components, such as CPU, memory, storage and network, while experiments are performed in order to compare the different power models used by the simulation frameworks. The following cloud simulation frameworks are considered: CloudSched, CloudSim, DCSim, GDCSim, GreenCloud and iCanCloud.

Extracting and exploiting the flexibility of electric demand has been shown to reduce the needs of network upgrades and generation capacity increases. Demand Response (DR) in considered as one of the few available solutions for accessing the untapped energy potential of small and medium customers. Over the past decade, rigorous research has produced significant results in optimally dispatching DR in an attempt to maximize flexibility extraction. However, the vast majority of works assumes a “happy path” scenario in which DR requests are always successfully completed. Hence, there is a large gap in the literature that fails to account for non-deterministic factors that manifest in practical deployments, e.g., the stochasticity of end-user behavior that can drastically influence the DR's outcomes. Investing on that notion, a novel, distributed, multi-agent system (MAS) that aggregates consumers and prosumers and handles automatically OpenADR-compliant DR requests is introduced, following virtual power plant (VPP) principles. Agents of the proposed MAS are able to service DR events originating from a higher level, e.g., Aggregators or Utilities, and optimally dispatch them to their assigned customers. The proposed framework ensures 100% DR success rate, compared to conventional methods, by not only optimally exploiting aggregated flexibility through a combination of clustering and optimisation engines, but also through a dynamic, bi-directional DR matchmaking process that can mitigate observed deviations both internally (intra), as well as, externally (inter) in real-time. Via experimentation, we demonstrate the framework's efficiency in ensuring technical DR fault-tolerance along with its ability to deliver savings of up to 3 orders of magnitude to Aggregators and the customers serving the DR requests.

Deterioration modelling of ageing phenomena on materials is an actively researched topic in computer graphics and vision, with a wide range of applications in domains such as cultural heritage, game programming, material science and virtual reality. As a result significant progress has been accomplished and existing methods are able to produce visually pleasing results that appear realistic. However, there is a very limited connection to comprehensive measurements that actually capture the ageing process of a material. This paper focuses on this gap, aiming to provide a link between physical measurements and deterioration modelling. Based on extensive measurements of texture and surface geometry of artificially aged reference materials, a Deep Learning (DL) framework is proposed that models spatio-temporal variations on the 3D surface geometry and the 2D colour–image appearance. Concretely, the problem of material degradation over time is formulated as an 2D/3D material-to-material translation problem, where the goal is, given an input material and a target degradation time, to output the degraded material at that time. At the core of the method lies a modified conditional Generative Adversarial Network (cGAN), which maps input materials to degraded materials over time. In order to train and deploy the proposed cGAN model, proper data parameterization and augmentation steps are introduced. As shown through extensive experimentation on real data coming from materials commonly found in artwork and from actual artworks, the proposed approach produces high quality results.

In this paper we propose a model for accurate traffic prediction under both normal and abnormal conditions. The model is based on the identification of the traffic patterns shown under both normal and abnormal conditions using the density-based clustering algorithm DBSCAN, and the use of different prediction models for each separate cluster that represents a traffic pattern. The k- Nearest Neighbor and the Support Vector Regression algorithms from the machine learning field and the ARIMA model from time series analysis were trained and tested. Preliminary experimental results indicate that the proposed model outperforms typical traffic prediction models from the relevant literature in terms of prediction accuracy under both normal and abnormal conditions.

In this paper, we focus on the building performance assessment using big data and visual analytics techniques driven by building occupancy. Building occupancy is a paramount factor in building performance, specifically lighting, plug loads and HVAC equipment utilization. Extrapolation of patterns from big data sets, which consist of building information, energy consumption, environmental measurements and namely occupancy information, is a powerful analysis technique to extract useful semantic information about building performance. To this end, visual analytics techniques are exploited to visualize them in a compact and comprehensive way taking into account properties of human cognition, perception and sense making. Visual Analytics facilitates the detailed spatiotemporal analysis building performance in terms of occupancy comfort, building performance and energy consumption and exploits innovative data mining techniques and mechanisms to allow analysts to detect patterns and crucial point that are difficult to be detected otherwise, thus assisting them to further optimize the building’s operation. The presented tool has been tested on real data information acquired from a building located at southern Europe demonstrating its effectiveness and its usability for building managers.

Being a widely adapted and acknowledged practice for the representation of inter- and intra-dependent information streams, network graphs are nowadays growing vast in both size and complexity, due to the rapid expansion of sources, types, and amounts of produced data. In this context, the efficient processing of the big amounts of information, also known as Big Data forms a major challenge for both the research community and a wide variety of industrial sectors, involving security, health and financial applications. Serving these emerging needs, the current paper presents a Graph Analytics based Platform (GAP) that implements a top-down approach for the facilitation of Data Mining processes through the incorporation of state-of-the-art techniques, like behavioural clustering, interactive visualizations, multi-objective optimization, etc. The applicability of this platform is validated on 2 istinct real-world use cases, which can be considered as characteristic examples of modern Big Data problems, due to the vast amount of information they deal with. In particular, (i) the root cause analysis of a Denial of Service attack in the network of a mobile operator and (ii) the early detection of an emerging event or a hot topic in social media communities. In order to address the large volume of the data, the proposed application starts with an aggregated overview of the whole network and allows the operator to gradually focus on smaller sets of data, using different levels of abstraction. The proposed platform offers differentiation between different user behaviors that enable the analyst to obtain insight on the network’s operation and to extract the meaningful information in an effortless manner. Dynamic hypothesis formulation techniques exploited by graph traversing and pattern mining, enable the analyst to set concrete network-related hypotheses, and validate or reject them accordingly.

Automated real-time occupancy monitoring in buildings plays an important role in increasing energy efficiency and provides facility managers with useful information about the usage of different spaces. In this paper, a novel approach is proposed for estimating real-time occupancy in buildings, based on Conditional Random Field probabilistic models, utilizing data from different sensor types. Three different types of occupancy information are considered: presence/absence, actual number of occupants and occupancy density. The proposed occupancy estimation method has been applied to four spaces with different characteristics in a real-life testbed environment. Experimental results revealed that the proposed method yielded high accuracy for different sensor combinations in all tested configurations regarding the occupancy granularity and the space type, and outperformed the Hidden Markov Model based method.

The fourth industrial revolution and the digitisation of supply chains have led companies to realize that the adoption of Industry 4.0/Internet of Things (IoT) solutions creates opportunities for more sustainable management. The sustainable management of scrap metal is a challenging task for all the organisations that participate in the supply chain and especially for scrap metal producers and waste management companies. Although metals are considered to be infinitely recycled, scrap metal management is often inefficient due to several factors including collection processes, communication and marketplace limitations, which have significant environmental and economic impacts. The purpose of this paper is to provide evidence of the impact of an IoT solution on the sustainable supply chain management (SSCM) performance. A case study from a scrap metal producer that operates in the lift industry and a waste management company is presented, in order to illustrate how the deployment of a state-of-the-art industry 4.0 solution has the potential to improve sustainability both in the firm level and in the supply chain level. Direct benefits of the introduced solution are the automation of monitoring and negotiation procedures for the produced scrap metal. Indirectly, the proposed solution is beneficial in terms of CO2 emissions’ reduction, resources availability and response time optimization. The results validate the framework for assessing SSCM for Industry 4.0 developed by Manavalan and Jayakrishna (2019) and demonstrate that Ιndustry 4.0 solutions have the potential to improve, among others, the economic, environmental and social sustainability in supply chain management. The present study contributes to the literature by bridging the gap between theoretical developments and real-world cases in the fields of industry 4.0 and SSCM. Managerial implications, limitations and future research opportunities are also provided.

This paper presents a generic methodology for fault forecasting or prognosis in industrial equipment. Particularly, this technique regards training some unsupervised machine learning model using high amount of historical process data from such equipment as input and stop data as reference. The goal is to correlate as strongly as possible the anomalies found by the model in the process data with upcoming faults, according to a forecasting horizon. In this way, the outlier detection model serves for fault forecasting. In this work the pre-processing and automatic feature selection phases, which are of significant importance, are also described. Such trained model would be useful for an industrial operator if executed in real time, based on online process data, since potential anomaly alerts raised by the model could enable predictive maintenance. This method has been applied to real industrial data related to aluminium and plastic production. The experimental results, with Matthews Correlation Coefficient up to 0.73 for the binary classification problem formulated to evaluate forecasting, provide strong evidence that machine learning models are capable of successfully forecasting upcoming faults before their occurrence, despite the general difficulty to find useful information in the process data for fault forecasting.

The concept of modeling the behavior of industrial processes is of great importance as it describes the possible states of equipment used in large industries, which once damaged, it usually costs both in time and money. In this paper, we propose a data-driven methodology for depicting three distinct states of a chemical reactor, (1) normal, (2) warning, (3) alert, by using machine learning techniques. A method for predicting the classification of data input, assists in prevention (early prognosis) of possible malfunctions. This method uses a combined linear trend analysis of the involved data which form the warning state of the reactor where the pre-incident conditions are fulfilled. Afterwards, it checks the possibility of the subsequent input to be classified in the alert state which is an indication that the reactor’s active equipment, such as heating resistance, will start malfunctioning. The objective of the three main steps of the proposed methodology are: first, to reveal the number of clusters based on past data, second to train normal, warning and alert behavior-models and validate them and third to test them as well as verify the accuracy of linear trend analysis. The proposed methodology is based on the analysis of real data sets​ derived from the automation system of a chemical process located at CERTH/CPERI in order to identify real-life models for prognostic behavior for malfunction prevention. This approach is especially suitable for modern industrial systems that follow Industry 4.0 principles. The results reveal a robust modeling of the reactor’s behavior with accuracy reaching 88,94%.

The prediction of upcoming events in industrial processes has been a long-standing research goal since it enables optimization of manufacturing parameters, planning of equipment maintenance and more importantly prediction and eventually prevention of defects. While existing approaches have accomplished substantial progress, they are mostly limited to processing of one dimensional signals or require parameter tuning to model environmental parameters. In this paper, we propose an alternative approach based on deep neural networks that simulates changes in the 3D structure of a monitored object in a batch based on previous 3D measurements. In particular, we propose an architecture based on 3D Convolutional Neural Networks (3DCNN) in order to model the geometric variations in manufacturing parameters and predict upcoming events related to sub-optimal performance. We validate our framework on a microelectronics use-case using the recently published PCB scans dataset where we simulate changes on the shape and volume of glue deposited on an Liquid Crystal Polymer (LCP) substrate before the attachment of integrated circuits (IC). Experimental evaluation examines the impact of different choices in the cost function during training and shows that the proposed method can be efficiently used for defect prediction., 21 pages, 5 figures

In this paper a building occupancy prediction method is presented, which is based on the spatio-temporal analysis of historical data (occupancy modelling) and further relies heavily on current contextual information, being therefore suitable for providing real-time prediction. Two different algorithmic approaches are proposed, based on Markov models, revealing how context awareness adds the capability of rapidly adjusting to current conditions and capturing unexpected events, as opposed to capturing only typical occupancy fluctuation expected on a regular basis. Both proposed approaches are evaluated against accurate real-life data collected from a tertiary building, achieving notable results which outperform currently used methods.

A real-time, multi-camera incident detection system for indoor environments is presented in this paper. The paper focuses on the detection of fall incidents while it highlights the leverage that such a system can provide to the human resources department of a shop-floor especially referring to the maintenance procedures. The proposed detection method extracts features that characterize a falling person’s trajectory, like vertical velocity and area variance, while the fall is described by Hidden Markov Models (HMM). The system utilizes only privacy preserving sensors. Experimental results illustrate its efficiency.

Virtual User Models (VUMs) can be a valuable tool for accessibility and ergonomic evaluation of designs in simulation environments. As increasing the accessibility of a design is usually translated into additional costs and increased development time, the need for specifying the percentage of population for which the design will be accessible is crucial. This paper addresses the development of VUMs representing specific groups of people with disabilities. In order to create such VUMs, we need to know the functional limitations, i.e. disability parameters, caused by each disability and their variability over the population. Measurements were obtained from 90 subjects with motor disabilities and were analyzed using both parametric and nonparametric regression methods as well as a proposed hybrid regression method able to handle small sample sizes. Validation results showed that in most cases the proposed regression analysis can produce valid estimations on the variability of each disability parameter.

This paper presents an extensive study on prehension-based dynamic features and their use for biometric purposes. The term prehension describes the combined movement of reaching, grasping and manipulating objects. The motivation behind the proposed study derives from both previous works related to the human physiology and human motion, as well as from the intuitive assumption that different body types and different characters would produce distinguishable, and thus valuable for biometric verification, activity-related traits. A novel approach for analyzing such movements is presented herein, based on the generation of an activity related manifold, the Activity hyper-Surface. The authentication capacity of the extracted features on the activity hyper-surface is evaluated in terms of their relative entropy and their mutual information within a complete framework targeting user verification. Experimental results on two datasets of 29 real subjects each and a third one of 100 virtual subjects show that the introduced concept constitutes a promising approach in the field of biometric recognition. HighlightsWe justify the validity of a prehension movement as biometric trait.We propose Activity hyper-Surfaces as descriptors for the arm movements.We analyze and combine Activity Curves of finger movements for completeness.Relative Entropy & Mutual Information based algorithm for dimensionality reduction.We evaluate our system on two real medium sized datasets and a large synthetic one.

IP video traffic is forecast to be 82% of all IP traffic by 2022. Traditionally, content distribution networks (CDN) were used extensively to meet quality of service levels for IP video services. To handle the dramatic growth in video traffic, CDN operators are migrating their infrastructure to the cloud and fog in order to leverage its greater availability and flexibility. For hyperscale deployments, energy consumption, cache placement, and resource availability can be analyzed using simulation in order to improve resource utilization and performance. Recently, a discrete-time simulator for modelling hierarchical virtual CDNs (vCDNs) was proposed with reduced memory requirements and increased performance using multi-core systems to cater for the scale and complexity of these networks. The first iteration of this discrete-time simulator featured a number of limitations impacting accuracy and applicability: it supports only tree-based topology structures, the results are computed per level, and requests of the same content differ only in time duration. In this paper, we present an improved simulation framework that (a) supports graph-based network topologies, (b) requests have been reconstituted for differentiation of requirements, and (c) statistics are now computed per site and network metrics per link, improving granularity and parallel performance. Moreover, we also propose a two phase optimization scheme that makes use of simulation outputs to guide the search for optimal cache placements. In order to evaluate our proposal, we simulate a vCDN network based on real traces obtained from the BT vCDN infrastructure, and analyze performance and scalability aspects.

Health data exchange is a major challenge due to the sensitive information and the privacy issues entailed. Considering the European context, in which health data must be exchanged between different European Union (EU) Member States, each having a different national regulatory framework as well as different national healthcare structures, the challenge appears even greater. Europe has tried to address this challenge via the epSOS ("Smart Open Services for European Patients") project in 2008, a European large-scale pilot on cross-border sharing of specific health data and services. The adoption of the framework is an ongoing activity, with most Member States planning its implementation by 2020. Yet, this framework is quite generic and leaves a wide space to each EU Member State regarding the definition of roles, processes, workflows and especially the specific integration with the National Infrastructures for eHealth. The aim of this paper is to present the current landscape of the evolving eHealth infrastructure for cross-border health data exchange in Europe, as a result of past and ongoing initiatives, and illustrate challenges, open issues and limitations through a specific case study describing how Italy is approaching its adoption and accommodates the identified barriers. To this end, the paper discusses ethical, regulatory and organizational issues, also focusing on technical aspects, such as interoperability and cybersecurity. Regarding cybersecurity aspects per se, we present the approach of the KONFIDO EU-funded project, which aims to reinforce trust and security in European cross-border health data exchange by leveraging novel approaches and cutting-edge technologies, such as homomorphic encryption, photonic Physical Unclonable Functions (p-PUF), a Security Information and Event Management (SIEM) system, and blockchain-based auditing. In particular, we explain how KONFIDO will test its outcomes through a dedicated pilot based on a realistic scenario, in which Italy is involved in health data exchange with other European countries. ispartof: JOURNAL OF BIOMEDICAL INFORMATICS vol:94 ispartof: location:United States status: published

In this study we examined the effects of intermittent short-duration Galvanic Vestibular Stimulation (GVS) during a multisensory perturbation of posture in young and elderly adults. Twelve young (24.91 +/- 6.44 years) and eleven elderly (74.8 +/- 6.42 years) participants stood upright under two task conditions: (a) quiet standing and (b) standing while receiving pseudo-randomly presented bipolar 2 s GVS pulses. In both conditions, sensory reweighting was evoked by visual surround oscillations (20 cm, 0.3 Hz) and Achilles tendon vibration (3 mm, 80 Hz), concurrently delivered during the middle 60 s of standing. Intermittent GVS decreased the excessive postural sway induced by the concurrent visual and proprioceptive perturbation in young but not in elderly participants. It is suggested that GVS increases sensory reliance on the vestibular system while elderly adults are less able to exploit this stimulation in order to reduce the destabilizing effect of the multisensory perturbation on their posture. (C) 2013 Elsevier Ireland Ltd. All rights reserved.

The use of spatial (geographic) information is becoming ever more central and pervasive in today’s internet society but the most of it is currently inaccessible to visually impaired users. However, access in visual maps is severely restricted to visually impaired and people with blindness, due to their inability to interpret graphical information. Thus, alternative ways of a map’s presentation have to be explored, in order to enforce the accessibility of maps. Multiple types of sensory perception like touch and hearing may work as a substitute of vision for the exploration of maps. The use of multimodal virtual environments seems to be a promising alternative for people with visual impairments. The present paper introduces a tool for automatic multimodal map generation having haptic and audio feedback using OpenStreetMap data. For a desired map area, an elevation map is being automatically generated and can be explored by touch, using a haptic device. A sonification and a text-to-speech (TTS) mechanism provide also audio navigation information during the haptic exploration of the map.

This paper presents novel subject-dependent biosignal features, with a view towards increasing the effectiveness of automatic psychological stress detection. The features proposed in this work focus on suppressing between-subject variability that typically appears in biosignals like the skin conductance (SC) and the electrocardiogram (ECG), and degrades the performance of relevant emotion recognition (ER) systems. For this purpose, the proposed features employ filtering of input signals, on the basis of ''rest signatures'' calculated from each subject's baseline recordings. These signatures are biosignal transformations capable to express each individual's baseline deviation from signal templates, which would ideally be applied during rest. The proposed subject-dependent features, extracted from SC and ECG modalities, were found capable to significantly increase automatic stress detection accuracy over a multi-subject (N=24) data set, collected through an experiment of natural stress induction. They provided accuracy at the level of 95%, significantly improved to the respective result (86.05%) taken from common SC and ECG features that have been typically used in the past. They appeared also similarly effective in automatic frustration detection over a further dataset. The results of the present work indicate that the proposed subject-dependent features, can lead to significant advances in the performance of future relevant ER systems.

We investigated the acquisition and transfer of a visually guided voluntary postural sway (PS) task that was practiced with the provision of either continuous or end point visual cues. Forty healthy adults were randomly assigned in to one of the four groups that practiced sway using different combinations of target and performance feedback. Participants were asked to voluntarily sway in the sagittal plane at a pre-set frequency (0.23 Hz) by matching the force exerted on a dual force platform to a visual target. Baseline, post-test, transfer and retention (24 h later) tests required performance of the PS task paced by a metronome. Continuous target cues resulted in greater accuracy at the peaks but at the cost of increasing movement intermittency. End-point cues on the other hand, produced more stable sway patterns but target overshooting. These adaptations differently generalized in the auditory-driven workspace, as reflected by more stable sway patterns for the groups practicing with end target cues and an enhancement of the ankle stiffening strategy for the groups practicing with continuous targets. It is suggested that the types of visual cues available during visually driven PS have a strong influence not only on the acquisition of this task but also on its generalization to the audio-motor workspace.

Despite its growing age, the JPEG is among the most popular choices as a standard compression scheme for continuous-tone still images. In this paper, a novel technique is proposed to alleviate the blocking artifacts that usually occur in JPEG coded images especially at low bit rates. The proposed algorithm consists of two stages: firstly, the AC coefficients are estimated based on their observed probability distribution and secondly, a postprocessing scheme is applied for blockiness removal, consisting of a region classification algorithm and a spatial adaptive filtering. Experimental results demonstrate the efficiency of the proposed method.

In the present paper a face recognition technique is developed based on depth and colour information. The main objective of the paper is to evaluate three different approaches (colour, depth, combination of colour and depth) for face recognition and quantify the contribution of depth. The proposed face recognition technique is based on the implementation of the principal component analysis algorithm and the extraction of depth and colour eigenfaces. Experimental results show significant gains attained with the addition of depth information.

Multiview image sequence processing has been the focus of considerable attention in recent literature. This paper presents an efficient technique for object-based rigid and non-rigid 3D motion estimation, applicable to problems occurring in multiview image sequence coding applications. More specifically, a neural network is formed for the estimation of the rigid 3D motion of each object in the scene, using initially estimated 2D motion vectors corresponding to each camera view. Non-linear error minimization techniques are adopted for neural network weight update. Furthermore, a novel technique is also proposed for the estimation of the local non-rigid deformations, based on the multiview camera geometry. Experimental results using both stereoscopic and trinocular camera setups illustrate and evaluate the proposed scheme.

This paper describes a procedure for model-based analysis and coding of both left and right channels of a stereoscopic image sequence. The proposed scheme starts with a hierarchical dynamic programming technique for matching across the epipolar line for efficient disparity/depth estimation. Foreground/background segmentation is initially based on depth estimation and is improved using motion and luminance information. The model is initialised by the adaptation of a wireframe model to the consistent depth information. Robust classification techniques are then used to obtain an articulated description of the foreground of the scene (head, neck, shoulders). The object articulation procedure is based on a novel scheme for the segmentation of the rigid 3D motion fields of the triangle patches of the 3D model object. Spatial neighbourhood constraints are used to improve the reliability of the original triangle motion estimation. The motion estimation and motion field segmentation procedures are repeated iteratively until a satisfactory object articulation emerges. The rigid 3D motion is then re-computed for each sub-object and finally, a novel technique is used to estimate flexible motion of the nodes of the wireframe from the rigid 3D motion vectors computed for the wireframe triangles containing each specific node. The performance of the resulting analysis and compression method is evaluated experimentally.

This paper describes a procedure for model-based coding of all channels of a multiview image sequence. The 3D model is initialised by accurate adaptation of a 2D wireframe model to the foreground object of one of the views. The rigid 3D motion is estimated for each triangle, and spatial homogeneity neighbourhood constraints are used to improve the reliability of the estimation efficiency and to smooth the motion field produced. A novel technique is used to estimate flexible motion of the nodes of the wireframe from the rigid 3D motion vectors of the wireframe triangles containing each node. Kalman filtering is used to track both rigid 3D motion of each triangle and flexible deformation of each node of the wireframe. The performance of the resulting 3D flexible motion estimation method is evaluated experimentally.

The transmission of the large store of information contained in 3D medical data sets through limited capacity channels, is a critical procedure in many telemedicine applications. In this paper, techniques are presented for the compression of visualisations of 3D image data for efficient storage and transmission. Methods are first presented for the transmission of the 3D surface of the objects using contour following methods. Alternatively, the visualisation at the receiver may be based on a series of depth maps corresponding to some motion of the object, specified by the medical observer. Depth maps may be transmitted by using depth map motion compensated prediction. Alternatively, a wire-mesh model of the depth map may be formed and transmitted by encoding the motion of its nodes. All these methods are used for the transmission of the 3D image with visualisation carried out at the receiver. Methods are also developed for efficient transmission of the images visualised at the encoder site. These methods allow remote interactive manipulation (rotation, translation, zoom) of the 3D objects, and may be implemented even if the receiver is a relatively simple and inexpensive workstation or a simple monitor. In all above cases, the coding of binocular views of the 3D scene is examined and recommendations are made for the implementation of coders of stereo views of 3D medical data.

Filters are determined which achieve optimal pyramidal decomposition of vector signals by minimising the variance of the errors between successive pyramid levels. Further, the filters are also determined which minimise the error produced when only one of the channels is retained from a multi-channel perfect reconstruction filter bank. The effect of either decomposition is to ensure that the lower-resolution image produced by the pyramid or the primary subband, bears maximum resemblance to the input image. This property is needed in applications requiring hierarchical (scalable, progressive) coding of vector fields. The noiseless case is examined first. Given arbitrary filters in one of the analysis or synthesis stages of the pyramid or the filter bank, the optimal corresponding synthesis, respectively analysis filters, are determined. Further, the globally optimal pairs of analysis and synthesis filters are determined. It is seen that under noiseless or lossless transmission conditions, the two above decomposition methods are optimised by identical families of analysis and synthesis filters. This is not the case if it is assumed that additive transmission noise corrupts the downsampled signal prior to the synthesis stage. This noise may represent transmission noise or the effect of quantisation between the analysis and synthesis stages. Given arbitrary analysis filters and arbitrary noise characteristics, the optimal synthesis filters are then determined for each decomposition method. The results are initially presented for the binary case and subsequently generalised to the case of M -factor pyramids and M -channel filter banks.

In the present paper techniques are examined for the coding of the depth map and disparity fields for stereo or multiview image communication applications. It is assumed that both the left and right channels of the multiview image sequence are coded using block- or object-based methods. A dynamic programming algorithm is used to estimate a disparity field between each stereo image pair. Depth is then estimated and occlusions are optionally detected, based on the estimated disparity fields. Spatial interpolation techniques are examined based on the disparity/depth information and the detection of occluded regions using either stereoscopic or trinocular camera configurations. It is seen that the presence of a third camera at the transmitter site improves the estimation of disparities, the detection of occlusions and the accuracy of the resulting spatial interpolation at the receiver. Various disparity field and depth map coding techniques are then proposed and evaluated, with emphasis given to the quality of the resulting intermediate images at the receiver site. Block-based and wireframe modeling techniques are examined for the coding of isolated depth or disparity map information. Further, 2D and 3D motion compensation techniques are evaluated for the coding of sequences of depth or disparity maps. The motion fields needed may be available as a byproduct of block-based or object-based coding of the intensity images. Experimental results are given for the evaluation of the performance of the proposed coding and spatial interpolation methods.

A novel motion compensation algorithm for the coding of 3D image sequences is presented. The temporal correlation existing between two successive in time 3D data sets is exploited using a 3D cube matching algorithm (CMA). In order to cope with more complex motions and to adapt compensation to the high activity regions, a 3D warping-based motion estimation technique is introduced. Simulation results are given for the 3D medical image sequence “Beating Heart”.

A method that performs 3-D motion estimation for stereoscopic image sequences is presented. The 2-D motion of each object observed in one of the two views is modelled using a 3-D motion model involving a translation and a rotation. The estimation of model parameters is performed in two steps: a linear step involving 2-D vectors that are initially estimated using block matching techniques followed by a non-linear step involving displaced frame difference minimization. The regions where the 3-D model is applied are identified using a motion-based split and merge technique. Furthermore, an extension of the 3-D motion estimation method that uses a single 3-D motion model to describe the apparent 2-D motion in both channels is examined. These 3-D motion estimation methods are then integrated in a stereoscopic interframe coding scheme. A hybrid coder using block-based coding as a fall-back mode in cases where 3-D motion estimation fails, is proposed. Experimental results demonstrate the performance of the proposed coding scheme.

Multiresolution block matching methods for both monocular and stereoscopic image sequence coding are evaluated. These methods are seen to drastically reduce the amount of processing needed for block correspondence without seriously affecting the quality of the reconstructed images. The evaluation criteria are the prediction error and the speed of the algorithm for motion, disparity, and fused motion and disparity estimation, in comparison with the full search (exhaustive) method. A new method is also proposed based in multiresolution techniques, for efficient coding of the disparity or the displacement vector field.