Unveiling deceit: A computer vision-based approach to detect lies from facial expressions

• Main Authors: Isaac, Elizabeth, Thoduparambil, Pristy Paul, Rohitha, M. Raghu, Ansalam, Anuvindha, Akshay, T. P.
• Format: Conference Proceeding
• Language: English
• Subjects: Aspect ratio; Centroids; Computer vision; Eye movements; Heart rate; Lie detectors; Pupils; Real time
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0230446

Polygraph tests monitor heart rate and pulse for truth determination but aren’t fully reliable. Face based deception detection systems are noninvasive and are more convenient. The proposed system uses Computer Vision to monitor eye blink rate, gaze direction, lip compression, and pupil dilation to determine truth accuracy in real-time. The number of frames with a threshold value lower than the Eye-Aspect-Ratio (EAR) constant determines the eye blink count. The pupil coordinates are used to track the eye gaze direction. The eye centroid to iris contour maximum distance tracks pupil dilation. Lip compression is determined by measuring the distance between the mean of upper and lower lip landmarks. The weighted sum method is used to find the lie index from the obtained features. The true positive rate (TPR) and false positive rate (FPR) at various threshold values is calculated and an ROC (Receiver Operating Characteristic) curve is plotted using these values. The point on the curve with the highest TPR and lowest FPR determines the optimal threshold for the lie detection system. This method allows customization of the system’s performance to meet specific requirements.