Experimental Evaluation of a Visual Detection Model

Abstract

A fixed-base simulation experiment was performed to gather visual air-to-ground target recognition performance data for comparison with predictions from the Autonetics Detection Model. Color motion picture imagery obtained during a low-altitude fright was used to simulate the observer's forward view. Observer performance was measured in terms of probability and range of correct target recognition. The Autonetics Detection Model incorporates parameters related to the target, the environment, and the observer. In generating theoretical predictions from the model, values of all parameters were specified independently of the data obtained in the experiment. No curve fitting techniques were used to improve the fit between the empirical and theoretical curves. Results indicated a close relationship between the obtained performance data and the model predictions. A product-moment correlation of +0.53, significant at the 0,001 level, was obtained between the empirical and theoretical 50% recognition ranges. INTRODUCTIONSeveral attempts to develop mathematical models of air-to-ground target recognition performance have recently been made (Gilmour and Emerson, 1965;Franklin and Whittenburg, 1965; and Ornstein, Brainard, and Bishop, 196 1). Two major problems exist in attempting to validate existing models. First, the lack of empirical performance data obtained under controlled conditions precludes comparisons of actual performances with theoretical model predictions over a substantial range of parameter values. Second, the large number of model parameters requires quantitative specification of many variables which are difficult to measure accurately.As a result, much model validation work has been forced into one of two paths. Either the model is simplified to the extent that it is no longer applicable to operational problems, or the data used to test the model arise from simplified experimental conditions. This paper presents a preliminary attempt to evaluate the Autonetics Detection Model. It was felt that the simulation method used for obtaining empirical data in this study would preserve sufficient "real-world'' detail in target and background parameters while retaining experimental control so that all model parameters could be quantitatively specified. THE AUTONETICS DETECTION MODEL Statement of the ModelThe objective of the visual detection and recognition model utilized by Autonetics is to furnish an estimate of the cumulative probability of having recognized a target by a certain point in the target approach. This estimate is to be made on the basis of measurable characteristics of the target and its background, the environment, and the observer. The model is largely descriptive rather than rational, in that the formulation of the model is fitted to the form of existing experimental data on several aspects of the search process without serious consideration of the underlying neurological mechanisms.Basically, the cumulative probability is obtained as a product of a series of single glimpse probabilities according ...

Locating Crash Sites in Simulated Air-to-Ground Visual Search

“…Consequently, the present study required observers t o search air-to-ground 16-mm color films of runs past several crash sites. Motion picture film has been used in other studies involved with eye-movements (Snyder, 1973) and the measurement of target size and contrast a t detection and recognition ranges (Greening and Wyman, 1970;Murphy, Overington, and Williams, 1965).…”

Air-to-Ground Target Acquisition Source Book: A Review of the Literature

“…PG generally increases because both contrast and target area increase as range-to-target decreases. The cumulative probability is then computed from: P cum One major validation study using a motion picture simulation has been conducted (Greening and Wyman, 1970). The results showed some agreement with the simulation data, a product-moment correlation of + .53, significant at the 0.001 level of confidence was obtained betwoen theoretical (model) and empirical (simulator) recognition ranges.…”

Mathematical Modeling of Air-to-Ground Target Acquisition

“…The effectiveness of search is input as the quantity, PL. Methods for determining PI, have been described (Greening, 1968;Greening and Wyman, 1970) but are not a part of the model as programmed.…”

“…Other modifications were made subsequently, based generally upon laboratory data. The entire Autonetics Model except the visual resolution function (resolution was limited by the simulator film) was subsequently validated against results of a cinematic simulation (Greening and Wyman, 1970), which showed good prediction without the necessity of adjusting any constants. Figure 16 shows the correlation between validation data a n d model prediction for all targets.…”