A neural network algorithm for longitudinal motion stereo

The dynamics of a discrete Hopfield neural network with multiple delays (HNNMDs) is studied by using a matrix inequality which is shown to be equivalent to the state transition equation of the HNNMDs network. Earlier work (2000) on discrete Hopfield neural networks showed that a parallel or serial mode of operation always leads to a limit cycle of period one or two for a skew or symmetric matrix, but they did not give an arbitrary weight matrix on how an updating operation might be needed to reach such a cycle. In this paper we present the existence conditions of limit cycles using matrix criteria in the HNNMDs network. For a network with an arbitrary weight matrix, the necessary and sufficient conditions for the existence of a limit cycle of period 1 and r are provided. The conditions for the existence of a special limit cycle of period 1 and 2 are also found. These results provide the foundation for many applications. A HNNMDs is said to have no stable state (fixed point) if it has a limit cycle of period 2 or more, which is stated in Theorem 5. A computer simulation demonstrates that the theoretical analysis in Theorem 5 is correct.

With the development of computer science and technology, the development of artificial intelli- gence advances rapidly in the field of image restoration. Based on the MATLAB platform, this paper constructs a kind of image restoration technology of artificial intelligence based on the discrete neural network and feedfor- ward network, and carries out simulation and contrast of the restoration process by the use of the bionic algo- rithm. Through the application of simulation restoration technology, this paper verifies that the discrete neural network has a good convergence and identification capability in the image restoration technology with a better effect than that of the feedforward network. The restoration technology based on the discrete neural network can provide a reliable mathematical model for this field.

Animals and humans rely on optic flow to navigate cluttered and unknown environments. While most previous studies have focused on how organisms achieve self-motion perception through optic flow information, biological neural networks for navigation based on optic flow remain unexplored. Here, we propose a biologically plausible neural network model for optic flow-based reactive navigation. The model incorporates a primary visual cortex, which is responsible for generating a cortical representation of the optic flow field; a higher-order cortex, which calculates the focus of expansion (FOE) of the optic flow field; and a cerebellum, which generates motor commands. A feedback inhibitory pathway from V1 layer VI to layer IV is introduced, enhancing heading sensitivity and enabling rapid adaptation in dynamic environments. To achieve precise obstacle localization, we propose a dual encoding strategy that combines optic flow with depth maps derived from the optic flow field, FOE, and control acceleration. This strategy mitigates distortions in depth estimation near the expansion center and ensures more reliable obstacle representation. The cerebellum outputs motor commands for heading direction and speed control based on the output of the visual cortex. Simulations and real-world experiments with an intelligent vehicle confirm that the proposed model enables collision-free navigation across diverse scenarios and outperforms classical optic flow balance strategies in complex environments. These findings demonstrate that biologically inspired neural networks provide a feasible solution for visual reactive navigation in autonomous agents.

Longitudinal motion stereo sinfers depth information from a forward or backward motion, and consequently is particularly useful in autonomous robot navigation applications. Most existing algorithms have some problems associated with the location of the FOE, and with the camera and surface orientations. These problems limit their applicability to real scenes. It is highly desirable to develop an efficient and practical algorithm to infer depth information from longitudinal motion. In this chapter, we present a neural network-based algorithm for longitudinal motion stereo. The algorithm allows the camera to move along its optical axis forward or backward, and requires no information on the FOE. It produces multiple dense disparity fields and recovers the depth map very efficiently.

Disadvantages of image proximity measure that is used in discrete neural network of adaptive resonance theory (ART) ART-1 are analyzed. The discrete network ART that uses the Hamming distance as an image proximity measure is proposed. Stable plastic Hamming neural network that is able to recognize the new information on its inputs is developed. Figs.: 2. Refs.: 9 titles. Keywords: neural network; adaptive resonance theory; Hamming distance; image proximity measure.

The article discusses the prerequisites and natural consequences of the control methods development in engineering systems: (1) a simple deviation and perturbation controller, (2) a fuzzy logic controller with a fuzzifier and a rule base, (3) a neural network controller for dynamically adjusting the coefficients of the corresponding links, (4) a discrete neural network controller with a neural approximator and controller. The experience gained by researchers and engineers since the first description of the principles of regulation in 1910 and the level of information technologies design, in particular the neural network method of machine learning and the enormous computing potential of computer devices, today can be integrated into a fundamentally new method of discrete neural network regulation.
 The review carried out in the article is aimed at identifying and demonstrating the significance of experimental and operational data, which must be properly structured and marked up at the stage of their collection and archiving. It is this approach that will allow us to quickly implement neural network controllers in engineering systems, since the most important stage for their creation is the process of learning and optimizing the architecture of neural networks.
 The principle of operation, advantages and disadvantages, the optimal stages of a neural network controller improvement based on two neural networks for the formation of a control strategy, taking into account the most probable state of the system at the next point in time, are given.

The difference in sensitivity to stereoscopic surfaces oriented horizontally or vertically (the stereoscopic orientation anisotropy) can be redescribed as a difference in sensitivity to shear or compression transformations that relate the binocular images. The present experiment was designed to test this by dissociating the image transformation from the orientation of the surface. Surfaces were presented in isolation or in the presence of a surrounding frame that formed step and gradient discontinuities in the disparity field. Without discontinuities, observers required considerably more time to discriminate between surfaces differing in compression than between those differing in shear, irrespective of surface orientation. Disparity discontinuities facilitated the perception of the disparity gradients; minimum stimulus durations were reduced by over an order of magnitude when the reference frame was present. These results support the hypothesis that the disparity field is decomposed into different primitives during the recovery of depth and surface structure.

This paper proposes an interval discrete wavelet neural network (IDWNN) model with learning algorithm, which can be used for the system modeling with uncertainty. Compared with the conventional interval BP neural network (IBPNN), IDWNN has both the time-frequency localization characteristics of wavelet analysis and the powerful nonlinear approximation ability, which can provide some guidelines for the network structure construction. The algorithms of the forward calculation and the gradient descent-based back propagation training of IDWNN are derived. The numerical simulation results show the advantages of IDWNN in network structure selection and convergence speed.

A piecewise linear equation is proposed as a method of analysis of mathematical models of neural networks. A symbolic representation of the dynamics in this equation is given as a directed graph on an N-dimensional hypercube. This provides a formal link with discrete neural networks such as the original Hopfield models. Analytic criteria are given to establish steady states and limit cycle oscillations independent of network dimension. Model networks that display multiple stable limit cycles and chaotic dynamics are discussed. The results show that such equations are a useful and efficient method of investigating the behavior of neural networks.

We have developed a universal chip for image processing by Discrete Time Cellular Neural Network (DTCNN) algorithm. We propose a ring-connected processor array architecture called pipelining DTCNN by which the convergence of output image is fast and done efficiently. In this paper, we introduce the hardware implementation of image halftoning by DTCNN with MOSIS, and the analysis of efficiency of pipelining DTCNN.

The dynamic behavior of discrete cellular neural networks (DCNN) with zero threshold value, which is strict, is mainly studied. For the DCNN with zero threshold value and no self-feedback, if a state is a fixed point, then a lot of unstable points are given, and under some conditions these unstable points can converge to the fixed point. In this paper, the properties of k-attractor of the DCNN are mainly studied, and some conditions are obtained under which the kattractor is a fixed point, and lots of unstable points are attracted to the fixed point. The results obtained here on k-attractor improve the results in the previous references.

New results on robust exponential stability for discrete recurrent neural networks with time-varying delays

The authors establish a qualitative theory for synchronous, discrete-time, Hopfield-type neural networks. Their objectives are accomplished in two phases. They analyze the class of neural networks considered and use the results to develop a synthesis procedure for them. The analysis utilizes techniques from the theory of large-scale interconnected dynamical systems to derive tests for the asymptotic stability of an equilibrium of the neural network. Estimates are obtained for the rate at which the trajectories of the network will converge from an initial condition to a final state. The authors utilize the stability tests as constraints to develop a design algorithm for content-addressable memories. The algorithm guarantees that each desired memory will be stored as an equilibrium and will be asymptotically stable. The applicability of the results is demonstrated for a 13-neuron and for an 81-neuron network. >

This paper describes Needles, an edge based stereo algorithm designed to take advantage of the smoothness of many textured surfaces. The correspondence problem is not addressed explicitly. Rather, a simple two stage process extracts surface position and orientation directly. Firstly local disparity histograms over a large range are constructed. Maxima in the histograms correspond to the possible surface depths. A Hough transform is used to fit a plane to the ambiguous disparity points close to the histogram maxima. This confirms and makes more precise the estimates of disparity obtained from the histograms. Local surface disparity and orientation are calculated from the best planar fit after all the histogram maxima (above a threshold) have been tried. This is an extension of an algorithm described in (Pollard 1985) which uses a Hough transform to find local surface orientation without explicit matching. In his algorithm pairs of possible matches vote for the disparity gradient between them. When all pairs have voted the winning disparity gradient (and hence, surface orientation) has the highest Hough accumulator value.

This article introduces a novel adaptive neural network compensator for feedforward compensation of external disturbances affecting a closed-loop system. The neural network scheme is posed so that a non-linear disturbance model estimate for a measurable disturbance can be adapted for rejection of the disturbance affecting a closed-loop system. The non-linear neural network approach has been particularly developed for ‘mobile’ applications where the adaptation algorithm has to remain simple. For that reason, the theoretical framework justifies a very simple least-mean-square approach suggested in a mobile hard disk drive context. This approach is generalised to a non-linear adaptive neural network (NN) compensation scheme. In addition, usual assumptions are relaxed, so that it is sufficient to model the disturbance model as a stable non-linear system avoiding strictly positive real assumptions. The output of the estimated disturbance model is assumed to be matched to the compensation signal for effectiveness, although for stability this is not necessary. Practical and simulation examples show different features of the adaptation algorithm. In a realistic hard disk drive simulation and a practical application, it is shown that a non-linear adaptive compensation scheme is required for non-linear disturbance compensation providing better performance at similar computational effort in comparison to well-established schemes.