A New Approach for Lie Detection Using Non-Linear and Dynamic Analysis of Video-Based Eye Movement

Abstract

Purpose: This study aimed to evaluate a lie-detection system by non-linear analysis of video-based eye movement. Materials and Methods: The physiological signals, as well as video-based eye movement in horizontal and vertical channels, were recorded based on a Control Question Test (CQT). The dynamics of eye movement signals were then analyzed by Recurrence Quantification Analysis (RQA). Statistical analysis was performed by ANOVA and Linear Discriminate Analysis (LDA). Results: In this study, 40 subjects participated. The statistical analysis results of vertical eye movement indicated that ENT measures increased significantly for relevant questions in comparison to other questions. Moreover, a significant increase was observed in all RQA parameters except Lmax and DET for horizontal eye movement. The results of LDA using psychophysiology features. The accuracy percentage of 78.4% and 81.86% were obtained for lie detection using physiological signals and optimal RQA parameters of video-based eye movements, respectively. Conclusion: The accuracy of lie detection by significant RQA parameters was more than the accuracy of physiological signals. So, the results of this study illustrate that the dynamic technique is well suited to analyze eye movement signals under stress and it could be recommended as a useful method in lie detection.