Dynamic Pursuit with a Bio-inspired Neural Model

Scientiae Mathematicae japonicae, 2006

We present the results of a computer vision system intended to describe in real time the trajectories of moving objects in a variety of speeds. The developed computing mechanism includes channel processing for instant position and velocity estimation, while trajectory plotting is made by combining direction and speed of movement. Retinal ganglion cell-like computational elements are randomly distributed thorough the input image, being direction-selective, allowing lateral interactions among them to correct peephole effects in movement analysis, presenting a variable degree of overlapping of their receptive fields (RF). We show results and errors of the working system when all parameters of cells are varied. One unexpected characteristic is that of the degradation of results in trajectory estimation when the size of receptive fields or the number of receptive fields, i.e. of computing cells, increase. In order to get good results for a variety of moving object speeds we have to reach a compromise between size and number of cells. This effect has interesting implications when translated back to the biological system which originally inspired the design of the computational method and could be used as a rationale to explain the distribution, morphological characteristics and number of movement detecting neurons. 1 General system description, implementation and results The study of biological basis of motion detection and analysis and the modelling and artificial implementation of those mechanisms has been a fruitful path of science in the last 60 years. When trying to implement an artificial vision system that has to work in very demanding conditions (e.g. real time, rapid changing visual environment etc) we found some of the biological concepts quite useful, thus is the case with receptive fields, overlapping, lateral interactions of computing units etc. [5], [6], [7]. However, some traditional approaches of visual computing (formal mathematical analysis, strict geometric patterns of neuron-like processors, selectivity of stimulae) have represented an extra restriction affecting to the overall performance of the system. After questioning these traditional approaches and including some "messy" characteristics of real biological systems such as random distribution of neuron-like processors, non homogeneity of neural architecture, channel processing, sudden failure of processing units and, in general, non deterministic behaviour of the system we found and show how reliability of motion analysis, computational cost and extraction of pure geometrical visual descriptors (size and position of moving objects) are improved in an implemented model. 1.1 Image acquisition and control of moving objects The system has been implemented on a Pentium TM PC type computer in which an acquisition and digitation board (IC-ASYNC, Imaging Technology TM) has been installed. This image board processes b/w

Journal of Neurophysiology, 2008

When we move forward the visual images on our retinas expand. Humans rely on the focus, or center, of this expansion to estimate their direction of self-motion or heading and, as long as the eyes are still, the retinal focus corresponds to the heading. However, smooth pursuit eye movements add visual motion to the expanding retinal image and displace the focus of expansion. In spite of this, humans accurately judge their heading during pursuit eye movements even though the retinal focus no longer corresponds to the heading. Recent studies in macaque suggest that correction for pursuit may occur in the dorsal aspect of the medial superior temporal area (MSTd); neurons in this area are tuned to the retinal position of the focus and they modify their tuning to partially compensate for the focus shift caused by pursuit. However, the question remains whether these neurons shift focus tuning more at faster pursuit speeds, to compensate for the larger focus shifts created by faster pursuit. To investigate this question, we recorded from 40 MSTd neurons while monkeys made pursuit eye movements at a range of speeds across simulated self-or object motion displays. We found that most MSTd neurons modify their focus tuning more at faster pursuit speeds, consistent with the idea that they encode heading and other motion parameters regardless of pursuit speed. Across the population, the median rate of compensation increase with pursuit speed was 51% as great as required for perfect compensation. We recorded from the same neurons in a simulated pursuit condition, in which gaze was fixed but the entire display counter-rotated to produce the same retinal image as during real pursuit. This condition yielded the result that retinal cues contribute to pursuit compensation; the rate of compensation increase was 30% of that required for accurate encoding of heading. The difference between these two conditions was significant (P Ͻ 0.05), indicating that extraretinal cues also contribute significantly. We found a systematic antialignment between preferred pursuit and preferred visual motion directions. Neurons may use this antialignment to combine retinal and extraretinal compensatory cues. These results indicate that many MSTd neurons compensate for pursuit velocity, pursuit direction as previously reported and pursuit speed, and further implicate MSTd as a critical stage in the computation of egomotion.

IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000

We have previously developed a neurodynamical model of motion segregation in cortical visual area V1 and MT of the dorsal stream. The model explains how motion ambiguities caused by the motion aperture problem can be solved for coherently moving objects of arbitrary size by means of cortical mechanisms. The major bottleneck in the development of a reliable biologically inspired technical system with real-time motion analysis capabilities based on this neural model is the amount of memory necessary for the representation of neural activation in velocity space. We propose a sparse coding framework for neural motion activity patterns and suggest a means by which initial activities are detected efficiently. We realize neural mechanisms such as shunting inhibition and feedback modulation in the sparse framework to implement an efficient algorithmic version of our neural model of cortical motion segregation. We demonstrate that the algorithm behaves similarly to the original neural model and is able to extract image motion from real world image sequences. Our investigation transfers a neuroscience model of cortical motion computation to achieve technologically demanding constraints such as real-time performance and hardware implementation. In addition, the proposed biologically inspired algorithm provides a tool for modeling investigations to achieve acceptable simulation time.

Proceedings International Conference on Information Technology: Coding and Computing,, 2001

In this paper, we examine the possibility that the spatiotemporal receptive field properties of v h a l cortical neurons can be understood in terms of a statistically efficient strategy for encoding natural time-varying images. It is believed that the sense of object motion and velocity are also related to these fields, as objects in natural scenes are represented by a sparse set of statistically independent components, such as edges. Currently, computational models of receptive fields consider only spatial components, and thus cannot account for time-varying sensory stimuli. In this paper, a model based on independent components analysis and cellular neural networks is proposed. We describe an artzjkial neural network that attempts to accurately reconsimct its spatiotemporal input data while simultaneously reducing the sutistical dependencies between its ourputs, as advocated ly the redundancy reduction principle. This approach extends existing models to incorporate temporal aspects of sequences of images of natural scenes.

Vision Research, 2005

During pursuit of a circularly moving target, the perceived movement of a second circularly moving target is altered. The perceived movement of the non-pursued target is different from both its real movement path and its retinal path. In the present paper this phenomenon is studied using a physiologically based neural network model. Simulation results were compared to psychophysical findings in human subjects. Model simulations enabled us to suggest an explanation for this phenomenon in terms of underlying physiological mechanisms and to estimate the contribution of the efferent eye-movement signal to the perceptual process.