Perceptual and behavioral functions of the brain are complexly linked to each other. Vision, for example, strongly relies upon top-down internal cognitive models that not only control perception and recognition of the external visual world but also related motor functionalities such as eye movements or instrumental visual search. It is possible to analyze experimentally several aspects of this top-down control from the observation of specific motor behaviors; clustering of target acquisitions is an important example. Metrics were developed to attempt to capture these aspects of the visual search process at a number of different levels; the metrics were designed to be either scale independent (two indices S p and S s ) or to define scale (the NK -functional curve). In order to estimate N , the number of clusters observed in a search scene, this NK -method was elaborated from the standard K -means algorithm. Two illustrative visual tasks, a quasi-natural search scene and picture viewing, were used in experiments to ascertain the effectiveness of these metrics. The results generally demonstrate the importance of top-down, normal vision and the applicability of the scanpath top-down theory for approaching the problems of visual search.

Independent and automatic image processing is a fundamental objective of the computer vision community. Understanding the role of the eye movement scan path in human vision is an important step toward the achievement of this objective. This top-down model of higher human vision is a new approach to bottom-up image processing algorithms and provides an important new metric and tool in computer vision. We have demonstrated that a small and manageable collection of image processing algorithms, experimentally selected and then combined together can serve in a task such as predicting human eye fixations identifying geological features. Thus, automatic picture analysis based upon human vision could be an essential element in planetary exploration.

Eye movements are an important aspect of human visual behavior. The temporal and space-variant nature of sampling a visual scene requires frequent attentional gaze shifts, saccades, to fixate onto different parts of an image. Experimental evidence suggests that fixations are often directed towards the most informative regions in the visual scene. We develop a model and its simulation that can select such regions based on prior knowledge of similar scenes. Having representations of scene categories as a probabilistic combination of hypothetical objects, i.e., prototypical regions with certain properties, it is possible to assess the likely contribution of each image region to the successive recognition process. Using conditional probabilities for each region given the scene category, the model can then predict its informative value and initiate a sequential spatial information-gathering algorithm, analogous to an eye movement saccade to a new fixation. This algorithm establishes the most likely scene category for a given image.

Eye movements (EMs) are an important aspect of human visual behavior. The temporal and space-variant nature of sampling a visual scene requires frequent attentional gaze shifts (saccades) to fixate onto different parts of an image. Fixations are often directed toward the most informative regions in the visual scene. We introduce a model and its simulation that can select such regions based on prior knowledge of similar scenes. Having representations of scenes as a probabilistic combination of regions with certain properties, it is possible to assess the likely contribution of each region in the successive recognition process. Using Bayesian conditional probabilities for each region given the scene category, the model can then predict the informative value of that region and initiate a spatial information-gathering algorithm analogous to an EM saccade to a new fixation.

Many machine vision applications, such as compression, pictorial database querying, and image understanding, often need to analyze in detail only a representative subset of the image, which may be arranged into sequences of loci called regions-of-interest (ROIs). We have investigated and developed a methodology that serves to automatically identify such a subset of aROIs (algorithmically detected ROIs) using different image processing algorithms (IPAs), and appropriate clustering procedures. In human perception, an internal representation directs top-down, context-dependent sequences of eye movements to fixate on similar sequences of hROIs (human identified ROIs). In the paper, we introduce our methodology and we compare aROIs with hROIs as a criterion for evaluating and selecting bottom-up, context-free algorithms. An application is finally discussed.

It is noted that a head-mounted display (HMD) system would be a useful teleoperator system display if it increased operator performance of the desired task but could degrade performance because of display update rate limitations and communication delays. 3-D tracking and pick-and-place tasks are simulated to characterize performance levels for a range of update rates and delays. How performance changes as the display is altered to maintain the update rate is also examined. The results of the update experiments indicate that performance levels degrade only slightly until the update rate is below 1 or 2 Hz. Attempts to maintain update rate by eliminating display features such as stereo and occlusion cues show that the presence of either cue alone is sufficient to maintain performance levels in the simple, short-term tasks considered. Delays also degrade performance in both manipulation tasks considered. These HMD performance problems are not lessened by the presentation of a predictive feedback display. Instead, the delay in the head movement tracking loop must be eliminated.>

Sequences of visual fixations, while looking at an object, are modeled as Markov processes, and statistical properties of such processes are derived by means of simulations. The sequences are also abstracted as character strings, and a quantitative method of measuring their similarity, based on minimum string editing cost (actually dissimilarity distance), is introduced. Interrelationships between the structure and size of the generating Markov matrices and the string editing distance shed light on the relative roles of deterministic and probabilistic processes in producing human visual scanpaths.>

Two types of depth cues contributed to the spatial perception of objects in scene: absolute cues, which theoretically provide information about the exact location of a point in space, and relative cues, which give depth information with respect to other points in space. The viewing parameters, such as the viewing distance or frustrum angle, can change the effect of these depth cues. The influence of disparity, object rotation, motion parallax, and the frustrum angle on a simulated telemanipulation task is examined. The results show that disparity is a powerful cue that can improve task performance. Object rotation and head motion parallax neither enhanced nor degraded performance. The frustrum angle affected the completion time but not error performance.>

A hierarchical telerobotic control system for experiments with model-based vision feedback is described. The human operator issues commands to a high-level controller which plans actions to be carried out by the low-level controller. A feedforward model of the robot's motion is used by the vision system to look for visual enhancements on the robot in their anticipated positions. A middle-level module is planned which will alter low-level control gains to maintain performance within error bounds. Results of experiments with the vision system and studies of control strategies for vision feedback are discussed.>

A simulation involving three types of telerobotic tasks that require information about the spatial position of objects is reported. This is similar to the results of psychophysical experiments examining the effect of blur on stereoacuity. It is suggested that other psychophysical experimental results could be used to predict operator performance for other telerobotic tasks. It is demonstrated that refractive errors in the helmet-mounted stereo display, system can affect performance in the three types of telerobotic tasks. The results of two sets of experiments indicate that monocular target blur of two diopters or more degrades stereo display performance to the level of monocular displays. This indicates that moderate levels of visual degradation that affect the operator's stereoacuity may eliminate the performance advantage of stereo displays.>

Two cooperative control schemes for telerobot visual displays are addressed. In the first scheme, on-the-screen visual enhancements such as reference lines indicating the vertical height of the robot hand, a stick figure model of the robot hand, and its projection on the horizontal grid plane are constructed by the interactive cooperation between the human operator and the telerobotic system, and then superimposed on the video screen. Experimental results with a five-degree-of-freedom robot and a frame grabber indicate that superimposition of visual enhancements on the video screen greatly improves telemanipulation task performance. In the second scheme, the position and orientation of an object on video screens are determined interactively; these then assist the telerobotic system in executing the human operator's task-level commands autonomously.>

It is argued that autonomous robotics is not yet feasible, and that instantaneous human control may be difficult because of communication delay and expensive because of labor costs. Therefore, it is argued, human supervisory control with cooperation between man and a partially autonomous distant robot is the most viable engineering solution. Model control of image processing enables efficient visual displays. Cooperative control in paired manipulators required rapid controller signal information to be passed over a cross-filter in a heuristic study.>

Position and rate control are the two common manual control modes in teleoperations. Human operator performance using the two modes is evaluated and compared. Simulated three-axis pick-and-place operations are used as the primary task for evaluation. First, ideal position and rate control are compared by considering several factors, such as joystick gain, joystick type, display mode, task, and manipulator work space size. Then the effects of the manipulator system dynamics are investigated by varying the natural frequency and speed limit. Experimental results show that ideal position control is superior to ideal rate control, regardless of joystick type or display mode, when the manipulation work space is small or comparable to the human operator's control space. Results also show that when the manipulator system is slow, the superiority of position control disappears. Position control is recommended for small-work-space telemanipulation tasks, while rate control is recommended for slow wide-work-space telemanipulation tasks.

A visual display system serves as an important man-machine interface for efficient teleoperations. However, careful consideration is necessary to display three-dimensional information on a two-dimensional screen effectively. A teleoperation simulator is constructed with a vector display system, joysticks, and a simulated cylindrical manipulator to evaluate various display conditions quantitatively. Pick-and-place tasks are performed, and mean completion times are used as a performance measure. Two experiments are performed. First, effects of variation of perspective parameters on human operator's pick-and-place performance with monoscopic perspective display are investigated. Then visual enhancements of monoscopic perspective display by adding a grid and reference lines are investigated and compared with visual enhancements of stereoscopic display. The results indicate that stereoscopic display does generally permit superior pick-and-place performance, while monoscopic display can allow equivalent performance when it is defined with appropriate perspective parameter values and provided with adequate visual enhancements. Mean completion time results of pick-and-place experiments for various display conditions shown are observed to be quite similar to the normalized rms error results of mannal tracking experiments reported previously.

An experimental telerobotics (TR) simulation is described suitable for studying human operator (HO) performance. Simple manipulator pick-and-place and tracking tasks allowed quantitative comparison of a number of calligraphic display viewing conditions. An enhanced perspective display was effective with a reference line from target to base, with or without a complex three-dimensional grid framing the view. This was true especially if geometrical display parameters such as azimuth (AZ) and elevation (EL) were arranged to be near optimal. Quantitative comparisons were made possible utilizing control performance measures such as root mean square error (rmse). There was a distinct preference for controlling the manipulator in end-effector Cartesian space for our primitive pick-and-place task, rather than controlling joint angles and then, via direct kinematics, the end-effector position. An introduced communication delay was found to produce decrease in performance. In considerable part, this difficulty could be compensated for by preview control information. That neurological control of normal human movement contains a sampled data period of 0.2 s may relate to this robustness of HO control to delay.

Optimal presentation of three-dimensional information on a two-dimensional display screen requires careful design of the projection to the display surface. Monoscopic perspective projection alone is usually not sufficient to represent three-dimensional spatial information. It can, however, be improved by the adjustment of perspective parameters and by geometric visual enhancements such as reference lines and a background grid. Stereoscopic display is another method of providing three-dimensional information to the human operator. Two experiments are performed with three-axis manual tracking tasks. The first experiment investigates the effects of perspective parameters on tracking performance. The second experiment investigates the effects of visual enhancements for both monoscopic and stereoscopic displays. Results indicate that, though stereoscopic displays do generally permit superior tracking performance, monoscopic displays can allow equivalent performance when they are defined with optimal perspective parameters and provided with adequate visual enhancements.

Four types of human movements are compared: eye saccades, horizontal head rotations, the twisting motion of the forearm about its long axis, and the horizontal rotation of the forearm about the axis of the elbow. Each is a rotation about a single axis of a load, driven by muscle-generated torques. In each case, earlier investigators have modeled the load as a second-order linear system and have identified the three parameters of the model. The authors' aim is to develop a hierarchy of descriptions of each kind of movement, by applying perturbation theory to the four second-order models. the amplitude, time, and maximum net torques are called to get a single-parameter equation for each movement system. Asymptotic series solutions are then developed for the response of each system to a step input, in terms of the single (small or large) parameter. The low-order terms of the expansions give the simplest approximations to the behavior of each system. These low-order approximate trajectories are used to compare the systems and to understand control strategies for different movements.

The theoretical basis of our experimental approach to information rate and capacity in a crayfish photoreceptor nerve channel is that used by Shannon [16]. Dispersion curves of conduction times were studied for pulses following short, medium, and long intervals. Although the means were equal (2.6 ms), the standard deviation in the case of pulses following short intervals was larger (0.29 ms) than for medium or long intervals (0.18 and 0.21 ms, respectively). A statistical description of nerve-fiber noise was found and used to compute the information rates (İ = 280, 140, and 80 bit/s) for various light intensities. By maximizing İ over all possible input interval distributions, the channel capacity was calculated to be 360 bit/s. Extending the channel to include the entire system defined from light stimulus to probabilistic behavioral output, the overall mutual information rate was estimated to be 0.04 bit/s, the discrepancy between this overall behavioral rate and the capacity of the nerve channel is noted.

A television pupillometer and an on-line computer were used to determine the static and dynamic characteristics of the pupillary system in the cat. The characteristics of sphincter and dilator mechanisms were measured with pulse rate modulated electrical stimuli to sympathetic and parasympathetic nerves and found to be similar to the light-driven pupillary response characteristic at high frequencies. Linear and nonlinear features of the response characteristics that were identified include retinal logarithmic operator, neural, neuromuscular, and mechanical saturation, high-frequency third-order dynamics of the motor component with a break frequency at 0.8 Hz, nonminimum phase transport delay of about 0.2 s, and response asymmetry to positive and negative steps of stimulation.