This paper discribes a study on the model of the human decision maker's behavior when predicting future states of discrete linear dynamic system driven by zero-mean Gaussian process.
In order to isolate only predicting behavior from human controller's characteristics, this paper considers a task where physiological limitations are insignificant because the time scale of the task is very slow.
It is found that the important information for a human operator is the two or three recent states of the history when ten points of system output are exhibited.
Futhermore, the multiple regression model expressing human prediction behavior is proposed. The accuracy of the proposed model is substantiated by checking the agreement between the model output and the operator's output by the correlation coefficient.

View full abstract

An experiment in which figures were observed through the central or the peripheral part of the visual field was conducted. The central part means a circular area containing the central fovea on the retina, and the remaining area is considered to be the peripheral visual field. The diameter of the central area was 3°, and figures subtended 15° at the subject's eye. Either the central or the peripheral part of the subject's visual field was covered throughout the observation. The subject, howerer, could pick up and observe any part of a figure by moving his eyes.
Three groups of unrecognized figures were found in the observation through the central visual field. In the observation through the peripheral visual field, a group of figures was found to give impressions which were different from those obtained in the ordinary observing condition. A model of the human figure recognition system was constructed on the basis of these results. The feature extractor of this model is a multilayered neural network whose characteristics can explain the difference in the function of figure perception between the central and the peripheral vision.

View full abstract

Now, inspections relying on human sensation are carried out finding defects and faults of industrial products in almost all corporations. And these inspections are much in need of workers and equipments, therefore, these are much expensive. On the other hand, in the industrial world it has been tried for automated machines to be substituted for such human inspections in view of cutting cost and human labor. By giving highly intelligent works to replace monotoneous inspection, we will look forward to recover humanity. Then we investigated about the present condition of and the demand for the visual inspections, which are the greater part of inspections relying on human senses.
The questionaire contains two part; 1) investigation of requirement to automated visual inspection, and 2) investigation of already automated examples. The questionaire were sent to 299 corporations, and the replies from 77 corporations were obtained. The replies were summarized by the subcommittee. Main results are as follows:
1) Visual inspections, which need many inspectors and are necessary to be automated are carried out finding defects in the semiconductor-warehars, errors in printing electric circuit patterns on IC boards, cracks and defects of tablets, and contents in cupsules.
2) Reasons for automating inspections are realization of objective inspection and improvement of their ability. As effects of automatic inspections, uniform quality, recovery of humanity released from monotoneous works, and reduction of cost are expected.
3) In already automated examples, techniques using laser, and photo-diode arrays as well as television technique are reported.
4) For visual inspections, a method of assessing necessity of automatic techniques is presented.

View full abstract