ME 224 Project: Lie Detector

June 6, 2003

Tim Fleck

Mike Gruener

Brian Halaburka

Chris Moskaites

Table of Contents

Summary……………………………………………………………………………………...…...2

Introduction to Polygraphs………………………………………………………………….……..2

Validity of Polygraphs……………………………………………………………………..…....2-3

Theory…………………………………………………………………………………………...3-4

Experimental Setup…………………………………………………………………….………..4-5

Procedure……………………………………………………………………………………...... 5-6

Logic…………………………………………………………………………………………….6-7

Data and Results………………………………………………………………………………...7-9

Difficulties……………………………………………………………………………………..8-12

Conclusions……………………………………………………………………………………....12

Appendix A: LABVIEW Programs………………………………………………………...... 13-15

Appendix B: References……………………………………………………………………...... 16

Summary:

For our final project for experimental engineering we will design, calibrate and implement a polygraph test. We will sample data from two sensors collecting two distinct types of data. The first sensor is a galvanic skin response sensor, which will measure the perspiration rate of the subject’s hand. The second sensor, which is an electrocardiogram, measures movement of the person who is being tested. After the detector is assembled, we will initially gather data from a group of volunteers. Their movement and perspiration will be recorded while they are asked a series of test questions. Afterwards they will be asked which questions they answered false. Based on these we will calibrate the data and determine how much of a change indicates a lie. From these determined values, a LABVIEW program will be assembled that will acquire future readings from subjects and compare these new values to the calibrated data. If the new readings are above our pre-determined level, the LABVIEW program will activate a light indicating that the subject is most likely telling a lie.

Introduction to Polygraphs:

A “lie detector” or polygraph instrument measures changes occurring in the body of a subject such as: heart rate, blood pressure, respiratory rate, electro-dermal activity, and arm and leg motions. These measurements are then compared to the normal levels of the subject. Polygraphs do not detect lies; however, they are designed to look for substantial involuntary changes in bodily rates, which occur in a person's body when that person is subjected to stress, such as the stress associated with deception.

Validity of Polygraphs:

Years of study have shown that people generally exhibit certain characteristics when they are lying. These studies have produced the modern polygraph. However, how do we know that these studies are accurate for all people? The American Polygraph Association states that it “has a compendium of over 80 research projects, involving 6,380 examinations. Researchers conducted 12 studies of the validity of field examinations, following 2,174 field examinations providing an average accuracy of 98%. Researchers conducted 11 studies involving the reliability of independent analyses of 1,609 sets of charts from field examinations confirmed by independent evidence providing an average accuracy of 92%.” Most errors occur with inexperienced polygraph examiners. One misconception with the accuracy figures presented by the American Polygraph Association is the fact that the numbers do not take inconclusive results into account. Thus a 98% accuracy reading does not mean that polygraphs are able to determine 98% of all questions correct, only 98% of the questions it is able to provide an answer for. Although the claims of nearly perfect accuracy for the modern polygraph are impressive, United States courts rarely allow them admissible. The inadmissibility of polygraphs in courts is mainly due to questions of inaccuracy. Most people are still uncertain about how emotions such as being nervous, scared, or embraced will affect the monitor. Other questions arise from subjects being able to manipulate the given rates being tested. Many websites and books are available on how to fool a polygraph. Some of the methods suggested are sedatives, antiperspirant on fingertips, tacks placed in shoes (to give the subject pain after each question), and biting the tongue, lip, or cheek. These countermeasures may not even go into the accuracy levels provided by the American Polygraph Association because the measures may produce inconclusive results, not errors. Although measures can be made to stop some countermeasures, some may be hard to detect. The validity of our detector is very inaccurate because of our design. We have fewer sensors than a normal polygraph and our sensors are likely to be less accurate. Due to these reasons, our lie detector is merely a primitive model that will portray how a real polygraph works.

Theory:

The idea behind a polygraph exam is a simple concept, but difficult to execute. When a person is being deceptive, the machine will notice that certain physiological activities are varying from their normal patterns. These “activities” are monitored by connecting various tubes and wires to specific points on a test subject’s body which will be monitored by a host of external sensors. The polygraph then ideally detects these physiological changes when a person is engaging in deceptive behavior. Then a trained examiner (also known as a forensic psycho physiologist) monitors the amount the activities fluctuate from their normal values. In general the physiological activities the polygraph monitors are respiratory rate, blood pressure/heart rate, galvanic skin resistance, and arm and leg motions.

Actual Polygraph (analog)

Respiratory rate: Two air filled rubber tubes are fastened around the subject’s abdomen and chest. When the muscles in these two bodily regions expand, the air in the fastened tube is displaced by a certain amount. This air then interacts with a device in the tube called a bellows, which contracts as the tubes expand. This bellows device is connected to a mechanical arm which itself is fastened to writing device (typically a pen) that marks on a moving sheet of paper every time the subject takes a breath.

Blood pressure/heart rate: An electrocardiogram is a painless and useful test that measures the electrical activity of the heart. Electrodes are placed on the chest, arms, and legs to monitor the heart’s rate and rhythm. An EKG is usually used for general health care and can help determine a variety of problems such as: disturbances of the heart's rhythm or rate, abnormalities in the axis, the direction of the heart's electrical flow, an enlargement of the heart, and damage from a previous heart attack. Blood pressure can be measured with an ordinary hospital blood-pressure cuff placed around the upper arm of the person being tested. As the blood is pumped through an arm it creates a sound, which causes a pressure change in the tubes. These tubes are also connected to a bellows, which again is connected to a pen that moves along with these disturbances.

Galvanic skin resistance: This measures the sweat being perspired by your fingertips. The fingertips were chosen because of their high pore to surface area density. On average when a person is placed under stress more sweat is excreted. Two fingerplates (called galvanometers) are attached to two of the subject’s fingers, these galvanometers then measure the ability of the skin to conduct electricity. Electricity is more easily conducted when the skin is moist (as with sweat) than when it is dry.

Arm and leg motion: Theoretically when a person is engaging in deceptive practices they will make more random bodily movements than normal. Thereby comparing the normal amount of bodily motion to the amount sensed when a question is asked a lie could possibly be detected.

Our Polygraph

Motion: We used the electrocardiogram (EKG) to measure bodily motions. Obviously an EKG can measure heart rate, but we found it too difficult to have LABVIEW take readings in real time. However, we did discover that the EKG could measure bodily motions, such as moving the arms and legs. Increased bodily motions are indications of increased stress and possible deception. By calibrating the LABVIEW program properly, we can determine when a subject is moving more than usual. If the signals cross a certain threshold value a green light goes on indicating that the person is probably engaged in deceptive behavior. (This is based on the assumption that when a person is being deceptive their movement rate will vary from their relaxed state.)

Galvanic skin resistance: Two of the test subject’s fingers are placed on two pads, which measure the moisture on the fingers. As the fingers become moister, electricity is conducted much easier. These values are then compared to the subject’s normal values and a predetermined threshold.

Experimental Setup:

Electrocardiogram

Originally, we set out to smooth this curve to indicate heart rate in real time. As will be discussed later, we had a variety of problems with this function. However, while attempting to read heart rate, we discovered that the EKG also senses when a person moves. Three sensors are placed on the subject, a positive, a negative, and a neutral. They are placed on both wrists and one ankle. The signal is sent to both an oscilloscope and to our LABVIEW program. LABVIEW is set up to read when a person is moving enough to make their signal pass a predetermined threshold.

Galvanic Skin Resistance

After several failed attempts at trying to make use of our GSR sensor, we decided to construct our own sensor that would measure conductivity of the skin. In order to do this, we used the principle that skin resistance decreases as it becomes moist. Dry skin has a resistance of about 1 million ohms, whereas the resistance of moist skin is reduced by a factor of ten or more. We used two resistors, both of 1,000,000, ohms to create a voltage divider. The probes were connected in parallel to one of the resistors, so as the resistance of the subjects’ skin changed, so would the output voltage. A decrease in the skin resistance would result in an increase in output voltage. The 100nF capacitor functions as a smoothing capacitor and removes the 50Hz induced mains hum that is found on a person's body.

We found that the maximum change in voltage output, from dry skin to very moist skin is around 0.3 V. Thus for our op-amp, we had a voltage gain of 10/0.3 = 33.33333. We chose resistors R1 and R2 to be 100,000 ohms and 3,000 ohms, respectively. This setup, with an offset voltage of around 1.5 V, gave us fairly accurate readings of skin moisture.

Procedure:

Professional Forensic Psych physiologist Procedure

Pretest – During this first hour or so, the examiner interviews the subject. During this time, the subject is not connected to the polygraph. The examiner gets to hear the subject’s side of the story. The examiner also visually sees how the subject responds to questions.

Design questions - The examiner designs questions that are specific to the issue under investigation and reviews these questions with the subject.
In-test – This is the phase when the polygraph exam is given. Some of the questions asked are relevant to the issue at hand. The other questions are asked as part of a control group. Control questions are generally very general so that the examiner knows the true answer to the question. Some example questions are: “Where do you live,” and
”Have you ever taken something that didn’t belong to you.” The subject will most likely tell the truth to the first question. The examiner can see what you bodily rates are at when the subject tells the truth. The second question is so vague that if any subject answers ‘no,’ the examiner can determine the person’s bodily rates during a lie.
Post-test - The examiner analyzes the data and makes a determination regarding whether the person was being truthful. If there are significant fluctuations that show up in the results, this may signal that the subject has been deceptive.

Our Procedure

Design questions – This phase is much like the one described above. Our group will design question to be answered by our subject. Unlike the professional examiner, we will not review the questions with our subject. This is mainly due to a shortage of time and would like to catch our subjects off guard.
In-test – The test is given. We will ask a group of control questions at the beginning of the exam to determine a control range. Based on this range, we will change certain variables to calibrate our system to the individual. This is slightly different than a professional examiner because we want to determine instantly if our subject is being truthful or not. Thus we must have our control questions at the beginning rather than interspersed throughout the exam.

Logic:

LABVIEW

-motion

Take initial readings visually from the waveform graph
Calibrate the high and low end of the threshold based on the initial readings
Every question after will be compared to the threshold
If a person’s motions go above the threshold, a LED goes on, indicating a lie

-galvanic skin response

Takes voltage readings of subject’s fingers until button is pushed
Then it computes and displays and average voltage for the given time period
Displays a waveform chart showing how much the subject’s voltage is increasing from the average as resistance decreases
If the in average increase goes above a certain predetermined value, a light goes on, indicating a lie

Project

Subject
Electrodermal Sensor / Motion Sensor
DAQ
New Data
LABVIEW Program/CPU / / Value Data Base
-Acceptable
-Not acceptable
Display Result

Data and Results:

Electrocardiogram

Below are graphs of our subjects being tested for their movements. The top is a graph when the person is not in motion and telling the truth. The bottom is a graph when the subject is in motion and lying.

Galvanic Skin Response

Below is a graph of our results from the galvanic skin response sensor. As can be seen, the subject was not lying because he/she was not above the predetermined threshold. If the curve went above 0.47 voltage increase, a lie would be indicated.

Difficulties:

Heart Rate Monitor

We made several attempts to measure the heart rate in order to use this as an additional input in our polygraph. We had a waveform on the oscilloscope, but we needed to convert the graphic form into a number of beats per minute. One of the most difficult aspects was updating the number in real time as we conducted the test.

To find the heart rate continuously, we needed to measure the number of beats within a specified time interval. We used a build array function in LABVIEW, and used both the sampling rate and the continual voltage reading as inputs. We set the array indices so that the array would change every millisecond to include the last twenty seconds of readings. This part of our effort was successful. However, we had to try several different methods of converting an array of voltage readings into a heart rate.

Our first method involved a series of mathematical operations on the array of voltage readings. First we differentiated the array, and then squared all of the values so that they would all be positive. Then we tried to smooth the curve to identify peaks, but we ran into problems. We were unable to find LABVIEW functions that would work with the data type that we needed, and we failed to convert the data into a more manageable type. We were able to find extrema on the curve, but noise from the signal prevented us from utilizing this data.

Our next effort involved MATLAB scripting. The fast Fourier transformation (fft) function in MATLAB is able to perform a discrete Fourier transformation on an array. With a series of algebraic and matrix manipulations, one can start with an array and come up with a frequency. In initial tests, it proved successful in isolating a sine curve from noise with over 5 times the amplitude of the sinusoid, and calculated the frequency to within 0.1%. However, when we applied this code into LABVIEW, it failed to provide reliable results. Occasionally it would produce a value that seemed accurate, but it would just as often give that same value off by a factor of 2. Most of the time there was no output, leading us to believe that the extensive MATLAB code, which performs several complex matrix operations in each step, was unable to keep up with the 1 kHz sampling frequency.

We next attempted to use some of the similar built-in LABVIEW functions. We tried several functions from the waveform and signal processing menus, such as variation from the mean and number of peaks. However, these were ultimately unsuccessful, and it is likely that they also lagged behind the readings, which were taken with a high sampling frequency.

Galvanic Skin Response

We bought a commercial perspiration sensor, the GSR2 Biofeedback Relaxation System from Bio-Medical Instruments, Inc. This device measures perspiration on the fingers and detects the resistance of two fingers. The resistance decreases as perspiration increases. The GSR2 emits an audible sound, and increases pitch with decreasing resistance. Thus a higher pitch implies the subject is more likely to be lying.

A higher pitch corresponds to a higher frequency electronic signal. We opened up the device and found the wires in the circuit board that led to the speaker. We soldered two external wires to these leads and hooked them up to our circuit board. Our intent was to send these signals into Labview, and use some of the waveform analysis tools to determine the frequency. After calibrating the data for each subject, we could establish a cutoff frequency that would indicate that the subject is lying.