Stelios C. A. Thomopoulos

Sensor integration and data fusion

The problem of sensor integration and data fusion is addressed. We consider the problem of combining information from diversified sources in a coherent fashion. We assume that information from various sensors may be available in different forms at the fusion. For example, data from infrared (IR) sensors may be combined with range radar (RR) data and further combined with visual images. In each case, data and information from different sensors are presented in a different format which may not be directly compatible for all sensors. Part of the available information may be in the form of attributes and part in the form of dynamical measurements. A generalized evidence processing theory and an architecture for sensor integration and data fusion that accommodates diversified sources of information are presented. Data (or, more generically, information) fusion may take place at different levels, such as the level of dynamics, the level of attributes, and the level of evidence. The common and different aspects of fusion at the different levels are investigated and several practical examples of real world data fusion problems are discussed.

Neural network implementation of the shortest path algorithm for traffic routing in communication networks

Summary form only given, as follows. A neural network computation algorithm is introduced to solve the optimal traffic routing in a general N-node communication network. The algorithm chooses multilink paths for node-to-node traffic which minimize a certain cost function (e.g. expected delay). Unlike the algorithm introduced earlier in this area, the knowledge about the number of links (hops) between each origin-destination pair is not required by the algorithm, therefore it can be applied to a more general network. The neural network structure for implementing the algorithm is a modification of the one used by the traveling salesman algorithm. Computer simulations in a nine-node grid network show that the algorithm performs well.<<ETX>>A neural network computation algorithm is introduced to solve for the optimal traffic routing in a general N-node communication network. The algorithm chooses multilink paths for node-to-node traffic which minimize a certain cost function. Unlike the algorithm introduced earlier in this area, knowledge of the number of links between each origin-destination pair is not required by the algorithm, therefore it can be applied to variable-length path routing problems. The neural network structure for implementing the algorithm is a modified form of the one used by the traveling salesman algorithm. Computer simulation in a nine- and sixteen-node grid network showed that the algorithm performs extremely well in single and multiple paths.<<ETX>>

Distributed decision fusion in the presence of networking delays and channel errors

Abstract The effects of transmission delay and channel errors on the performance of a distributed sensor system are studied. In a network of distributed sensors at a given time instant, the decisions from some sensors may not be available at the fusion center owing to networking and transmission delays. Assuming that the fusion center has to make a decision on the basis of the data from the rest of the sensors, provided that at least one peripheral decision has been received, it is shown that the optimal decision rule that maximizes the probability of detection for fixed probability of false alarm at the fusion center is the Neyman-Pearson test at the fusion center and the sensors as well. Furthermore, it is shown that, in the case of noisy channels, the decision made by each sensor depends on the reliability of the corresponding transmission channel. Moreover, the probability of false alarm at the fusion is restricted by the channel errors. For a given decision rule, the probability of any channel being in error must be kept at a certain level to achieve a desired probability of false alarm at the fusion. A suboptimal but computationally efficient algorithm is developed to solve for the sensor and fusion thresholds sequentially. Numerical results are provided to demonstrate the closeness of the solutions obtained by the suboptimal algorithm to the optimal solutions.

An iterative solution to the inverse kinematics of robotic manipulators

Abstract The solution to the inverse kinematics of robotic manipulators is important in the planning of trajectories to be executed by the manipulators in the task space. The problem of inverse kinematics of robotic manipulators involves the determination of a set of feasible joint angles which corresponds to a set of desired spatial coordinates in the task space. The set of desired spatial coordinates in the task space in turn represents the location of a desired point along a prescribed trajectory for the tip of a given manipulator. An iterative solution to the inverse kinematics is presented based on numerical methods for nonlinear algebraic equations. The necessary sets of spatial coordinates of a prescribed trajectory for the tip of both kinematically nonredundant and redundant robotic manipulators are determined using the algorithm. Simulation results from nonredundant and redundant manipulators are presented.

Novel control of an inverted pendulum

High frequency vertical oscillations of the pivot point of the inverted pendulum are investigated as a means of stabilizing the pendulum. Four cases are considered. The standard cart-pendulum is analyzed using optimal control techniques both with and without a vertical pivot oscillation. The stability of a simple inverted pendulum is investigated with a pure tone vertical displacement and then a gradient search is applied to determine a modulated vertical displacement to achieve asymptotic stability.

Nonlinear adaptive detection/estimation for single and multiple radar processing

A nonlinear adaptive detector/estimator is introduced for single and multiple radar data processing. The problem of target detection from returns of monostatic radar(s) is formulated as a nonlinear joint detection/estimation problem on the unknown parameters in the signal return. The problems of detecting the target and estimating its parameters are considered jointly. A bank of spatially and temporally localized nonlinear filters is used to estimate the a posteriori likelihood of the existence of the target in a given space-time resolution cell. Within a given cell, the localized filters are used to produce refined spatial estimates of the target parameters. A decision logic is used to decide on the existence of a target within any given resolution cell based on the a posteriori estimates obtained from the likelihood functions. Simulation results show excellent detection capabilities and excellent resolution in target parameter estimation for both single and multiple sensor data.<<ETX>>

Joint detection/estimation for modal order selection in Gaussian and nonGaussian noise

The authors present a general approach to determining the number of sinusoids present in measurements corrupted by additive white Gaussian and nonGaussian noise. The approach involves the simultaneous application of maximum a posteriori detection and nonlinear estimation using either the extended Kalman filter when the noise is Gaussian, or the extended high order filter when the noise is nonGaussian. The problem is formulated as a multiple hypothesis testing problem with assumed known a priori probabilities for each hypothesis. The advantage of the approach lies in the potential to accommodate time-varying as well as time-invariant parameters in the measurement model. Experimental evaluation of the approach demonstrates excellent performance in selecting the correct model order and estimating the system parameters even for signal-to-noise ratios as low as -5 dB.<<ETX>>

DIGNET: A self-organizing neural network for automatic pattern recognition and classification

A self-organizing artificial neural network (ANN) that exhibits deterministically reliable behavior to noise interference when the noise does not exceed a specified level of tolerance is presented. The complexity of the proposed ANN, in terms of neuron requirements versus stored patterns, increases linearly with the number of stored patterns and their dimensionality. The self-organization of the proposed ANN is based on the idea of competitive generation and elimination of attraction wells in the pattern space. The same neural network can be used both for pattern recognition and classification. The ANN has been tested successfully with pattern and signal recognition and classification paradigms.<<ETX>>

Comparative Study of Self-Organizing Neural Networks

A benchmark study of self-organizing neural network models is conducted. The comparison of advantages and disadvantages of unsupervised learning artificial neural networks are discussed. The unsupervised learning artificial neural networks discussed in this paper include adaptive resonance theory (ART2), DIGNET, self-organizing feature map, and learning vector quantization (LVQ). For the benchmark study of artificial neural network applications on data clustering and pattern recognition problems with additive gaussian noise, we compare the performance of the unsupervised learning systems, ART2 and DIGNET. Results of computer simulation show that both ART2 and DIGNET achieve good performance on pattern clustering, but DIGNET is faster in the learning process and has better results on the overall clustering performance.

Image restoration by inhomogeneous G-M field modeling and adaptive Kalman filtering

An inhomogeneous Gauss-Markov (G-M) model of the image field is used to enhance the visual quality of the restored image by Kalman filtering in the spatial domain. In the corresponding Kalman filter, the transition matrix in the predicting part is no longer a constant matrix as it was in previous homogeneous G-M modeling. Instead, it becomes a function of the spatial coordinates as well as the edge running parameter /spl theta/. The optimal estimate is a weighted sum of several Kalman filter estimates with each Kalman filter operating in parallel with a separate known value of /spl theta/. The experimental results are compared with those obtained in the homogeneous G-M modeling.<<ETX>>