BieColl - Bielefeld eCollections
Not a member yet
1004 research outputs found
Sort by
Tracking Nuclear Material at Low Frame Rate and Numerous False Detections
Get PDFIn Nuclear Safeguards, surveillance cameras monitor the correct processing of nuclear material. Nuclear inspectors are faced with tens of thousands of images to review, of which less than 1% is significant. Besides the reduction of the standard two-frame differencing filter, we further limit the image set to review by tracking on the distribution of image time-stamps. Traditional visual tracking cannot be applied, owing to the low frame rate, and the need for compatibility with the standard change detection algorithm. Our algorithm is based on a HMM model of the nuclear process, and handles multiple flasks and observations available only when the flasks are moved. The model makes use of descriptive statistics of the durations of processing stages to refine the HMM predictions
Advanced metric adaptation in Generalized LVQ for classification of mass spectrometry data
Get PDFMetric adaptation constitutes a powerful approach to improve the performance of prototype based classication schemes. We apply extensions of Generalized LVQ based on different adaptive distance measures in the domain of clinical proteomics. The Euclidean distance in GLVQ is extended by adaptive relevance vectors and matrices of global or local influence where training follows a stochastic gradient descent on an appropriate error function. We compare the performance of the resulting learning algorithms for the classification of high dimensional mass spectrometry data from cancer research. High prediction accuracies can be obtained by adapting full matrices of relevance factors in the distance measure in order to adjust the metric to the underlying data structure. The easy interpretability of the resulting models after training of relevance vectors allows to identify discriminative features in the original spectra
Indices to Evaluate Self-Organizing Maps for Structures
Get PDFSelf-Organizing Maps for Structures (SOM-SD) are neural networks models capable of processing structured data, such as sequences and trees. The evaluation of the encoding quality achieved by these maps should neither be measured only by the quantization error as in the standard SOM, which fails to capture the structural aspects, nor by other topology preserving indexes which are ill-defined for discrete structures. We propose new indexes for the evaluation of encoding quality which are customized to the structural nature of input data. These indexes are used to evaluate the quality of SOM-SDs trained on a benchmark dataset introduced earlier in. We show that the proposed indexes capture relevant structural features of the tree encoding additional to the statistical features of the training data labels
Modular Network SOM and Self-Organizing Homotopy Network as a Foundation for Brain-like Intelligence
Get PDFIn this paper, two generalizations of the SOM are introduced. The first of these extends the SOM to deal with more generalized classes of objects besides the vector dataset. This generalization is realized by employing modular networks instead of reference vector units and is thus called a modular network SOM (mnSOM). The second generalization involves the extension of the SOM from "map" to "homotopy", allowing the SOM to deal with a set of data distributions rather than a set of data vectors. The resulting architecture is called SOM^n, where each reference unit represents a tensor of rank n. These generalizations are expected to provide good platforms on which to build brain-like intelligence
SOM-based Peptide Prototyping for Mass Spectrometry Peak Intensity Prediction
Get PDFIn todays bioinformatics, Mass spectrometry (MS) is the key technique for the identification of proteins. A prediction of spectrum peak intensities from pre computed molecular features would pave the way to better understanding of spectrometry data and improved spectrum evaluation. We propose a neural network architecture of Local Linear Map (LLM)-type based on Self-Organizing Maps (SOMs) for peptide prototyping and learning locally tuned regression functions for peak intensity prediction in MALDI-TOF mass spectra. We obtain results comparable to those obtained by nu-Support Vector Regression and show how the SOM learning architecture provides a basis for peptide feature profiling and visualisation
3D Scene Segmentation and Object Tracking in Multiocular Image Sequences
Get PDFIn this contribution we describe a vision-based system for the 3D detection and tracking of moving persons and objects in complex scenes. A 3D point cloud of the scene is extracted by a combined stereo technique consisting of a correlation-based block-matching approach and a spacetime stereo approach based on spatio-temporally local intensity modelling. Hence, the result of stereo analysis is a 3D point cloud attributed with motion information. For localising persons and objects in the scene the point cloud is segmented into meaningful clusters by applying a hierarchical clustering algorithm, using velocity information as an additional discrimination criterion. Initial object hypotheses are obtained by partitioning the observed scene with cylinders, including the tracking results of the previous frame. Multidimensional unconstrained nonlinear minimisation is then applied to refine the position, velocity and size of the initial cylinder in the scene, such that neighbouring clusters with similar velocity vectors are grouped to form a compact object. A particle filter is applied to select hypotheses which generate consistent trajectories. The described system is evaluated based on a tabletop sequence and several real-world sequences acquired in an industrial production environment, based on manually obtained ground truth data. We find that even in the presence of moving objects closely neighbouring the person, all objects are detected and tracked in a robust and stable manner. The average tracking accuracy is of the order of several percent of the distance to the scene
A Real-time Visual Attention System Using Integral Images
Get PDFSystems which simulate human visual attention are suited to quickly find regions of interest in images and are an interesting preprocessing method for a variety of applications. However, the scale-invariant computation of features in several feature dimensions is still too time consuming to be applied to video streams at frame rate which is necessary many practical applications. As a consequence, current implementations of attention systems often make compromises between the accuracy and speed of computing a focus of attention in order to reduce the computation time. In this paper, we present a method for achieving fast, real-time capable system performance with high accuracy. The method involves smart feature computation techniques based on integral images. An experimental validation of the speed gain of the attention system VOCUS is provided, too. The real-time capability of the optimized VOCUS system has already been demonstrated in robotic applications
Adaptive Image Sampling and Windows Classification for On-board Pedestrian Detection
Get PDFOn-board pedestrian detection is in the frontier of the state-of-the-art since it implies processing outdoor scenarios from a mobile platform and searching for aspect-changing objects in cluttered urban environments. Most promising approaches include the development of classifiers based on feature selection and machine learning. However, they use a large number of features which compromises real-time. Thus, methods for running the classifiers in only a few image windows must be provided. In this paper we contribute in both aspects, proposing a camera pose estimation method for adaptive sparse image sampling, as well as a classifier for pedestrian detection based on Haar wavelets and edge orientation histograms as features and AdaBoost as learning machine. Both proposals are compared with relevant approaches in the literature, showing comparable results but reducing processing time by four for the sampling tasks and by ten for the classification one
Attention Based Auto Image Cropping
Get PDFMany images contain salient regions that are surrounded by too much uninteresting background material and are not as enlightening as a sensibly cropped version. The choice of the best picture window both at capture time and during subsequent processing is normally subjective and a wholly manual task. This paper proposes a method of automatically cropping visual material based upon a new measure of visual attention that reflects the informativeness of the image
Autonomic Computer Vision Systems
Get PDFMost computer vision systems perform well under controlled laboratory conditions, but require lengthy set up and "tuning" by experts when installed in new operating conditions. Unfortunately, for most real applications of computer vision, the operating conditions frequently change. These changes degrade system performance and can even cause complete system failure, requiring intervention by a trained engineer. The requirement for installation and frequent maintenance by highly trained experts seriously inhibits the commercial application of computer vision systems. In this talk we discuss ways in which autonomic computing can reduce the cost of installation and configuration, as well as enhance reliability, for practical computer vision systems. We begin by reviewing the origins of autonomic computing. We then describe the design of a computer vision system as a software component within a layered service architecture. We describe techniques for regulation of internal parameters, error detection and recovery, self description, and self configuration for vision systems. These methods will be illustrated with results from the IST projects FAME, CAVIAR and CHIL