Pranav Gopalabhatla, Lucas Ceurty, Avaneesh Joshi, Ashwin Manjrekar

Santa Cruz Works and Santa Cruz County Office of Education are proud to announce the second place winner for the Best High School Project at Cruz Hacks 2022: Re-LIEable. This team enjoys learning at Leigh High School.

Problem

Over the past years, our systems to prove if someone has been lying have been the same: the polygraph. In light of the events during covid like the riot at the Capital or the death of George Floyd or Kyle Rittenhouse and more, there has been no one brought to justice. With these peaceful protests like the BLM movement, African Americans are met with unnecessary violence that White Americans give. These minor problems are an undersized representation of the more significant problem in the world, where there are millions of crimes happening each day. People have been caught for their crimes but could get away from it by just passing a polygraph. Thus we made a more promising product that gave more accurate results to see if someone was lying. We can better understand whether someone is lying or telling the truth with our device.

Solution

Our solution is Re-LIEable. Re-LIEable is an interface that is able to tell if someone is lying based on their facial expressions. The machine learning models are able to calculate if a person is lying based on their blinking, sound, and expression detection. We used different machine learning models to find the highest success rate. There was an increase in improvement on a task that has monumental applications ranging from criminal investigations to national security and will overall uphold Justice in this country.

What it does

Our product works by detecting facial expressions through facial recognition. We use a webcam to detect their face and use sounds to see if their pitch changes. When a person asks questions, the web camera will see if their expression was different from before. There are no signs of grins, laughs, and the symmetry in face alignment changes. We compiled all of this and sent it through a single button to make the process easy and direct. With this, we have a history of the timestamps of when the person lied.

How we built it

We saw how easy it was to pass a polygraph while giving the wrong answers. So we wanted to build a more reliable product and have better results than the polygraph. So after numerous hours of hard work and going through challenges, we coded an entire app in python. We split the work by doing the making code to access the webcam and building a bounding rectangle to get a boundary around the person's face. With that, we added the estimated age of the person, their gender and, their expression to see if the changes during each question. With this, we are able to check if the person is lying or not.

Challenges we ran into

The main obstacle that we faced was making the machine learning algorithm work. We could not quite figure out how to do the database configurations but we figured it out after several hours of debugging. Also, the machine learning model had a lot of bugs and errors in it. However, we went through and spent time trying to solve them as a team before moving on to others parts of the project. Another problem was when trying to import node.js functions and libraries it wouldn’t properly integrate into the code due to the conflicting without libraries. However, we overcame these obstacles by working and collaborating together through verbal communication in Discord and Zoom.

Accomplishments that we're proud of

We were able to solve all the problems that we came across and we were able to problem solve as a team. Our user interfaces, in particular, turned out to be very successful in being both clean, atheistically pleasing, practical, and easy to both use and navigate. As a key component in our project, getting the Clarifai API to function properly with the application is another accomplishment that we are very proud of. Seeing as how the API took many attempts and trials to get working, it was definitely also one of the more challenging accomplishments.

What we learned

We learned that machine learning can be difficult but it is rewarding. We learned how to make an established website and incorporate 3d pictures to make the page aesthetically pleasing. We learned how to integrate the front and back end properly, which is one of the most crucial things to building a website. To add to that, we also learned how to feature elements that are more aesthetic for a better UX on top of all the necessary code.

What's next for Re-LIEable

We plan to make a more complex interface with more assets and additional data. We plan to make the product a startup. We believe that this product could lead to a better system of justice with telling if the person is guilty/lying often or not guilty/says the truth.