Knowledge engineering in the domain of visual behaviour of pilots

Abstract

Low level flight training carried out in flight simulators with computer generated image visual displays can lead to visual habit patterns (Comstock et al., 1987; Dixon et al., 1990) of the pilots, which may cause disorientation in real world flight (Haber, 1987). This study is an attempt to contribute to the improvement of visual simulation systems by means of knowledge about visual skills in different situations of flight. This knowledge can be taken account of in the design process of computer image generation so that there are elements of the visual scene, which pilots use as source of visual information by fixating them in special situations in real world flight (Kleiss, 1990; Kennedy et al., 1988). Situation in this context means the complete set of input information to the pilot at one point of time including the visual input of the out-of-window scene and the aircraft displays, the speech input from the copilot, the pilot’s knowledge about the mission plan and the current task and other inputs like mechanical cues. The pilot responds on each situation with certain actions including aircraft operations and eye movements. The eye movements have to be measured in a suitable experimental environment. The Universität der Bundeswehr München is carrying out a study for the development of such an environment. In order to investigate under the most realistic situations, the experiments should be performed in real flight. However, because of principal operational restrictions, this is not possible. Therefore, a video replay of a real flight’s visual scene is used in a research flight simulator. The paper presents the experimental concept, using the NAC-EMR 600 eye movement measuring system on pilots engaged in visual tasks in connection with the video replay. The data analysis is described. The variety of possible situations is represented by a number of feature vectors, allowing a numerical classification. The fixated scene elements can thereby be related to the corresponding situation classes. The visual behaviour is defined by the most significant correlations of fixated scene element classes and certain situation classes.