Research on neural network models for eye tracking and object fixation

Abstract

The paper presents a comparison of neural network models for eye tracking and object fixation. In this paper, event detection algorithms from different domains were studied: I-VT and I-DT, based on threshold values, a random forest model in the field of machine learning, and a CNN model in the field of deep learning. Their classification performance was evaluated using the same dataset for all methods. The relationship between programmers and algorithms was also discussed. The results showed the critical impact of threshold values on the classification of I-VT and I-DT algorithms. In this regard, finding the optimal threshold for threshold-based algorithms is a difficult task. RF and CNN algorithms outperform threshold-based algorithms in all performance indicators and are capable of classifying data into multiple classes. Various ABCs (event detection algorithms) have been tested and compared. Theoretical and experimental justifications for the results obtained have been given. Based on the results of the experiments, the best ABC for our tasks (event detection algorithms) has been selected. The means of implementation for creating the application are also described. The advantages and disadvantages of the selected solutions are described. The technical problems that have arisen have been solved. The architecture has been designed, and a program layout has been developed based on the required tasks.