Determining Response Credibility by Blink Count

摘要

There exist many forms of deception detection techniques such as Polygraph, Graphology but they are invasive and expert dependent system which can be sometimes easily thwarted by criminals. This project aims to utilize a physiological factor that is involuntary during lying to measure the response credibility of a suspect. One such factor is eye blink count and some studies proposed that blink rate increases during deception, whereas some studies proposed that during deception blink rate decreases. The objective of the current study is to analyze blink count by developing a computer vision application that is capable of detecting and counting eye blinks in video streams using facial landmarks to determine the credibility of response of various suspects while answering the interrogative questions. In this paper an experiment was conducted where 16 subjects were grouped into two and were asked to perform a mock crime and Interrogated where questions were based on the Concealed Information Test. The analysis shows that during critical questions directly related to crime scene, the suspect tends to have higher blink rate than normal whereas innocents tend to have lowered blink rate than under normal situation. This system is expected to be used in places such as Visa Interviews, Interrogations on Airports, petty criminal and civil case investigations in courts etc., where the credibility of the statements provided by the suspects is of great importance.