Refine
Year of publication
Document Type
- Conference Proceeding (179)
- Article (64)
- Part of a Book (16)
- Bachelor Thesis (6)
- Book (5)
- Report (5)
- Contribution to a Periodical (3)
- Doctoral Thesis (2)
- Master's Thesis (2)
- Part of Periodical (2)
Language
- English (287) (remove)
Is part of the Bibliography
- no (287)
Keywords
Given the success of convolutional neural networks (CNNs) during recent years in numerous object recognition tasks, it seems logical to further extend their applicability to the treatment of three-dimensional data such as point clouds provided by depth sensors. To this end, we present an approach exploiting the CNN’s ability of automated feature generation and combine it with a novel 3D feature computation technique, preserving local information contained in the data. Experiments are conducted on a large data set of 600.000 samples of hand postures obtained via ToF (time-of-flight) sensors from 20 different persons, after an extensive parameter search in order to optimize network structure. Generalization performance, measured by a leave-one-person-out scheme, exceeds that of any other method presented for this specific task, bringing the error for some persons down to 1.5 %.
This contribution presents a novel approach of utilizing Time-of-Flight (ToF) technology for mid-air hand gesture recognition on mobile devices. ToF sensors are capable of providing depth data at high frame rates independent of illumination making any kind of application possible for in- and outdoor situations. This comes at the cost of precision regarding depth measurements and comparatively low lateral resolution. We present a novel feature generation technique based on a rasterization of the point clouds which
realizes fixed-sized input making Deep Learning approaches applicable using Convolutional Neural Networks. In order to increase precision we introduce several methods to reduce noise and normalize the input to overcome difficulties in scaling. Backed by a large-scale database of about half
a million data samples taken from different individuals our
contribution shows how hand gesture recognition is realiz-
able on commodity tablets in real-time at frame rates of up to 17Hz. A leave-one out cross-validation experiment
demonstrates the feasibility of our approach with classification errors as low as 1,5% achieved persons unknown to the model.
To reduce the number of traffic accidents and to increase the drivers comfort, the thought of designing driver assistance systems rose in the past years. Principal problems are caused by having a moving observer (ego motion) in predominantly natural surroundings. In this paper we present a solution for a flexible architecture for a driver assistance system. The architecture can be subdivided into four different parts: the object-related analysis, the knowledge base, the behavior-based scene interpretation, and the behavior planning unit. The object-related analysis is fed with data by the sensors (e.g., vision, radar). The sensor data are preprocessed (flexible sensor fusion) and evaluated (saliency map) searching for object-related information (positions, types of objects, etc.). The knowledge base is represented by static and dynamic knowledge. It consists of a set of rules (e.g. , traffic rules, physical laws), additional information (i.e., GPS, lane-information) and it is implicitly used by algorithms in the system. The scene interpretation combines the information extracted by the object related analysis and inspects the information for contradictions. It is strongly connected to the behavior planning using only information needed for the actual task. In the scene interpretation consistent representations (i.e., bird's eye view) are organized and interpreted as well as a scene analysis is performed. The results of the scene interpretation are used for decision making in behavior planning, which is controlled by the actual task. The influence of behavior planning on the behavior of the guided vehicle is limited to advices as no mechanical control (e.g. , control of the steering angle) was implemented. An Intelligent Cruise Control (ICC) is shown as a spin-off for using this architecture.
We present a novel approach of distributing matrix multiplications among GPU-equipped nodes in a cluster system. In this context we discuss the induced challenges and possible solutions. Additionally we state an algorithm which outperforms optimized GPU BLAS libraries for small matrices. Furthermore we provide a novel theoretical model for distributing algorithms within homogeneous computation systems with multiple hierarchies. In the context of this model we develop an algorithm which can find the optimal distribution parameters for each involved subalgorithm. We provide a detailed analysis of the algorithms space and time complexities and justify its use with a structured evaluation within a small GPU-equipped Beowulf cluster.
A Large and Quick Induction Field Scanner for Examining the Interior of Extended Objects or Humans
(2017)
This study describes the techniques and signal properties of a large, powerful, and linear-scanning 1.5 MHz induction field scanner. The mechanical system is capable of quickly reading the volume of relative large objects, e.g., a test person. The general approach mirrors Magnetic Induction Tomography (MIT), but the details differ considerably from currently-described MIT systems: the setup is asymmetrical, and it operates in gradiometric modalities, either with coaxial excitation with destructive interference or with a single excitation loop and tilted receivers. Following this approach, the primary signals were almost completely nulled, and test objects' real or imaginary imprint was obtained directly. The coaxial gradiometer appeared advantageous: exposure to strong fields was reduced due to destructive interference. Meanwhile, the signals included enhanced components at higher spatial frequencies, thereby obtaining a gradually improved capability for localization. For robust signals, the excitation field can be powered towards the rated limits of human exposure to time-varying magnetic fields. Repeated measurements assessed the important signal integrity, which is affected by the scanner´s imperfections, particularly any motions or respiratory changes in living beings during or between repeated scans. The currently achieved and overall figure of merit for artifacts was 58 dB for inanimate test objects and 44 dB for a test person. Both numbers should be understood as worst case levels: a repeated scan with intermediate breathing and drift/dislocations requires 50 seconds, whereas a single measurement (with respiratory arrest) takes only about 5 seconds.
A light-weight real-time ap- plicable hand gesture recognition system for automotive applications
(2015)
We present a novel approach for improved hand-gesture recognition by a single time-of-flight(ToF) sensor in an automotive environment. As the sensor's lateral resolution is comparatively low, we employ a learning approach comprising multiple processing steps, including PCA-based cropping, the computation of robust point cloud descriptors and training of a Multilayer perceptron (MLP) on a large database of samples. A sophisticated temporal fusion technique boosts the overall robustness of recognition by taking into account data coming from previous classification steps. Overall results are very satisfactory when evaluated on a large benchmark set of ten different hand poses, especially when it comes to generalization on previously unknown persons.
In this paper we describe a session management system for setting up various collabora- tive classroom ,scenarios. The approach ,is addressing the additional workload ,of administrating classroom networks on the teacher, which is an important aspect for teachers' willingness to im- plement technology enhanced,learning in schools. The system facilitates preparation of classroom scenarios and the adhoc installation of networked collaborative sessions. We provided a graphical interface, which is usable for administration, monitoring, and for specification of a wide variety of different classroom ,situations with group work. The resulting graphical specifications are well suited to be re-used in the more formal learning design format IMS/LD; this is achieved by a auto- matable transformation of the scenarios to LD documents. Keywords: Collaborative classroom scenarios, lightweight classroom orchestration, learning de- sign, shared workspaces.
We present a novel hierarchical approach to multi-class classification which is generic in that it can be applied to different classification models (e.g., support vector machines, perceptrons), and makes no explicit assumptions about the probabilistic structure of the problem as it is usually done in multi-class classification. By adding a cascade of additional classifiers, each of which receives the previous classifier's output in addition to regular input data, the approach harnesses unused information that manifests itself in the form of, e.g., correlations between predicted classes. Using multilayer perceptrons as a classification model, we demonstrate the validity of this approach by testing it on a complex ten-class 3D gesture recognition task.