BieColl - Bielefeld eCollections
Not a member yet
1004 research outputs found
Sort by
Maximum-Likelihood Stereo Correspondence using Field Programmable Gate Arrays
Get PDFEstimation of depth within an imaged scene can be formulated as a stereo correspondence problem. Typical software approaches tend to be too slow for real time performance on high frame rate (>= 30fps) stereo acquisition systems. Hardware implementations of these same algorithms allow for parallelization, providing a marked improvement in performance. This paper will explore one such hardware implementation of a maximum-likelihood stereo correspondence algorithm on a Field Programmable Gate Array (FPGA). The proposed ""FastTrack"" hardware implementation is a first stage prototype that demonstrates comparable results to an equivalent software implementation. with the advantage of high-speed (eventually up to 200fps) stereo depth estimation
Pop-out and IOR in Static Scenes with Region Based Visual Attention
Get PDFThis paper proposes a novel approach to construct the saliency map by combining region-based maps of distinct features. The multiplication style feature fusion process in the natural visual attention is modelled as weighted average of the features under influence of the external top-down and the internal bottom-up inhibitions. The recently discovered aspect of feature-based inhibition is also included in the procedure of IOR along with the commonly implemented spatial and feature-map based inhibitions. Results obtained from the proposed method are compatible with the well known attention models but with the advantages of faster computation, direct usability of focus of attention in machine vision, and broader coverage of visually prominent objects
Detection of ambiguous patterns in a SOM based recognition system: application to handwritten numeral classification
Get PDFThis work presents a system for pattern recognition that combines a self-organising unsupervised technique (via a Kohonen-type SOM) with a bayesian strategy in order to classify input patterns from a given probability distribution and, at the same time, detect ambiguous cases and explain answers. We apply the system to the recognition of handwritten digits. This proposal is intended as an improvement of a model previously introduced by our group, consisting basically of a hybrid unsupervised, self-organising model, followed by a supervised stage. Experiments were carried out on the handwritten digit database of the Concordia University, which is generally accepted as one of the standards in most of the literature in the field
Label Propagation for Semi-Supervised Learning in Self-Organizing Maps
Get PDFSemi-supervised learning aims at discovering spatial structures in high-dimensional input spaces when insufficient background information about clusters is available. A particulary interesting approach is based on propagation of class labels through proximity graphs. The Self-Organizing Map itself can be seen as such a proximity graph that is suitable for label propagation. It turns out that Zhu's popular label propagation method can be regarded as a modification of the SOM's well known batch learning rule. In this paper, an approach for semi-supervised learning is presented. It is based on label propagation in trained Self-Organizing Maps. Furthermore, a simple yet powerful method for crucial parameter estimation is presented. The resulting clustering algorithm is tested on the fundamental clustering problem suite (FCPS)
Learning Vector Quantization: generalization ability and dynamics of competing prototypes
Get PDFLearning Vector Quantization (LVQ) are popular multi-class classification algorithms. Prototypes in an LVQ system represent the typical features of classes in the data. Frequently multiple prototypes are employed for a class to improve the representation of variations within the class and the generalization ability. In this paper, we investigate the dynamics of LVQ in an exact mathematical way, aiming at understanding the influence of the number of prototypes and their assignment to classes. The theory of on-line learning allows a mathematical description of the learning dynamics in model situations. We demonstrate using a system of three prototypes the different behaviors of LVQ systems of multiple prototype and single prototype class representation
Architecture and Tracking Algorithms for a Distributed Mobile Industrial AR System
Get PDFIn Augmented Reality applications, a 3D object is registered with a camera and visual augmentations of the object are rendered into the users field of view with a head mounted display. For correct rendering, the 3D pose of the users view w.r.t. the 3D object must be registered and tracked in realtime, which is a computational intensive task. This contribution describes a distributed system that allows to track the 3D camera pose and to render images on a light-weight mobile frontend user interface system. The frontend system is connected by WLAN to a backend server that takes over the computational burdon for realtime tracking. We describe the system architecture and the tracking algorithms of our system
Assisting persons with dementia during handwashing using a partially observable Markov decision process.
Get PDFThis paper presents a real-time system to assist a person with dementia wash their hands. Assistance is given in the form of verbal and/or visual prompts, or through the enlistment of a human caregiver's help. The system uses only video inputs, and combines a Bayesian sequential estimation framework for tracking hands and towel, with a decision theoretic framework for computing policies of action -- specifically a partially observable Markov decision process (POMDP). A key element of the system is the ability to estimate and adapt to user states, such as awareness, responsiveness and overall dementia level. We demonstrate the system in a set of simulation experiments, and we show examples of real-time interactions with actors
Automatic Analysis of Lens Distortions in Image Registration
Get PDFGeometric image registration by estimating homographies is an important processing step in a wide variety of computer vision applications. The 2D registration of two images does not require an explicit reconstruction of intrinsic or extrinsic camera parameters. However, correcting images for non-linear lens distortions is highly recommended. Unfortunately, standard calibration techniques are sometimes difficult to apply and reliable estimations of lens distortions can only rarely be obtained. In this paper we present a new technique for automatically detecting and categorising lens distortions in pairs of images by analysing registration results. The approach is based on a new metric for registration quality assessment and facilitates a PCA-based statistical model for classifying distortion effects. In doing so the overall importance for lens calibration and image corrections can be checked, and a measure for the efficiency of accordant correction steps is given
Face Alignment by 2.5D Active Appearance Model Optimized by Simplex
Get PDFIn this paper we propose an efficient algorithm to align the face in real time, based on Active Appearance Model (AAM) in 2.5D. The main objective is to make a robust, rapid and memory efficient application suitable for embedded systems, so they could align the pose rapidly by using less memory. Classical AAM is a high memory consumer algorithm, consequently transfer of this stored memory in an embedded system makes it a time consuming algorithm as well. Our 2.5D AAM is generated by taking 3D landmarks from frontal and profile view and 2D texture only from frontal view of the face image. Moreover we propose Nelder Mead Simplex technique for face search. It does not require large memory, thus becoming suitable for embedded systems by eliminating the excess memory and access time requirements. We illustrate 2.5D AAM optimized by Simplex for pose estimation and test it on three databases: M2VTS, synthetic images and webcam images. Results validate our combination of simplex and AAM in 2.5D
Implementation of an Affine-Covariant Feature Detector in Field-Programmable Gate Arrays
Get PDFThis article describes an FPGA-based implementation of the Harris-Affine feature detector introduced by Mikolajczyk and Schmid. The system is implemented on the Transmogrifier-4, a prototyping platform that includes four Altera Stratix S80 FPGAs and NTSC/VGA video interfaces. The system achieves a speed of 90-9000 times the speed of an equivalent software implementation, allowing it to process standard video (640 x 480 pixels) at 30 frames per second