A.V. Alekseev – Magister, Master, Laboratory Technician Medical Cybernetics, Department of Digital Technology, Voronezh State University
Ya.A. Тurovsky – Ph.D (Med.), Associate Professor, Head of Laboratory Technician Medical Cybernetics, Department of Medical Cybernetics Digital, Voronezh State University
E.V. Bogatikov – Ph.D. (Phys.-Math.), Associate Professor, Department of Semiconductor Physics and Mi-croelectronics, Voronezh State University
S.V. Boldyrev – Student, Computer Science Faculty, Voronezh State University
In this paper, we analyzed the possibilities of predicting the generation of commands for the eye-tracking interface in con-ditions of controlling the self-propelled chassis under the control of the camera installed on it. As empirical data were taken the results of 30 subjects, each of whom underwent an experiment three times learning to control the video-oculographic interface. As an effector device, a self-propelled chassis was used with a video camera attached to it, broadcasting the image to the monitor in front of the subject. As algorithms for forecasting commands generated by the user, artificial neural networks were used in the perceptron topology with one hidden layer, classification trees implemented on the basis of C @ RT and discriminant separation algorithms. The prognosis was made for a period of 2.5 s. For the prediction of the pupil position, linear regression was used. The forecast was carried out for a period of time in 2 seconds. Artificial neural networks showed the best result regarding the forecast of the commands. The accuracy of the forecast of the user's generation of certain commands, using three-layer perceptrons, with the number of hidden layer neurons up to 40, and classification trees realized by algorithms based on discriminant branching and C @ RT, does not depend on how well the user mastered the control with the video-doculographic Interface of the self-propelled chassis. The accuracy of the forecast for different commands is statistically significantly different, with the exception of the "back" and "right" commands, although it was not possible to establish a link between the sequence number of the arrival or its success with the forecast accuracy for any of the user's commands. Therefore, the available trained INS can be applied to the forecast, regardless of the stage of training of a particular user. At the same time, it is obvious that it is optimal to apply the forecast not to all the commands, but only to the part: "there is no command", "forward" and, possibly, "left". The "back" and "right" commands have a low forecast accuracy, which can lead to a large number of false positives of the system. The results of the regression analysis showed that the outcomes of the races and the sequence numbers of the races are not significant predictors for a successful prediction of the pupil's position dynamics. The greatest number of statistically significant coefficients for regression occurs in the previous second of the forecast. In this case, the previous value for the X coordinate in the first second is more important for the prediction on the same coordinate than the coordinates along the Y axis. A similar situation was for the Y-coordinate prognosis. A better prediction is provided with a prediction of the pupil position along the Y axis. The results obtained Allow to consider the forecast of generation of commands by the user on time intervals of the order of one or two seconds as a realizable opportunity to modify the management of external devices for a number of users.
- Venda V.F. Sistemy' gibridnogo intellekta: E'volyucziya, psixologiya, informatika. M.: Mashinostroenie. 1990. 448 s.
- Turovskij Ja.A., Kurgalin S.D. Oczenka skorosti raboty' nejrokomp'yuternogo interfejsa, realizovannogo s ispol'zovaniem gibridnogo intellekta // Biomediczinskaya radioe'lektronika. 2015. № 3. S. 61–70.
- Turovskij Ja.A., Kurgalin S.D., Alekseev A.V. Analiz dvizheniya glaz cheloveka pri upravlenii samoxodny'm shassi s ispol'zovaniem sistemy' videookulograficheskogo interfejsa // Sensorny'e sistemy'. 2017. T. 31. № 1. S. 51–58.
- Turovskij Ja.A., Kurgalin S.D., Vaxtin A.A., Borzunov S.V., Alekseev A.V. Informaczionny'e potoki pri formirovanii dopolnitel'nogo kanala obratnoj svyazi dlya interfejsa mozg-komp'yuter // Texnologii zhivy'x sistem. 2016. T. 13. № 7. S. 34–40.
- Kurgalin S.D., Alekseev A.V., Turovskij Ja.A. Komp'yuternaya sistema monitoringa dvizheniya glaz // V sb.: Informatika: problemy', metodologiya, texnologii: Materialy' XV Mezhdunar. nauch.-metod. konf. 2015. S. 178–180.
- Glancz S. Mediko-biologicheskaya statistika. M.: Praktika. 1998. 459 s.
- Runion R.P. Spravochnik po neparametricheskoj statistike: Sovremenny'j podxod: Per. s angl. E.Z. Demidenko. M.: Finansy' i statistika. 1982. 198 s.
- Coates A., Learning A.Y.Ng. Feature Representations with K-means. Stanford University. 2012.