Lie Detector

Lie Detector lets you monitor the heart rate and possible 'tells' of deception from any face, including live video calls or recordings.

Video demo and more info available here!

Lie Detector uses OpenCV and MediaPipe's Face Mesh to perform real-time detect of facial landmarks from video input. It also uses FER for mood detection. From there, relative differences are calculated to determine significant changes in specific facial movements from a person's baseline, including their:

Heart rate

Blink rate

Change in gaze

Hand covering face

Lip compression

Lie Detector can optionally include prompts based on a second video feed to better 'mirror' the original input.

Hit Q on the preview window to exit the resulting display frame, or CTRL+C at the terminal to close the Python process.

Optional flags:

--help - Display the below options

--input - Choose a camera, video file path, or screen dimensions in the form x y width height - defaults to device 0

--landmarks - Set to any value to draw overlayed facial and hand landmarks

--bpm - Set to any value to include a heart rate tracking chart

--flip - Set to any value to flip along the y-axis for a selfie view

--landmarks - Set to any value to draw detected body landmarks from MediaPipe

--record - Set to any value to write the output to a timestamped AVI recording in the current folder

--second - Secondary video input device for mirroring prompts (device number or path)

--ttl - Number of subsequent frames to display a tell; defaults to 30

Example usage:

python intercept.py -h - Show all argument options

python intercept.py --input 2 --landmarks 1 --flip 1 --record 1 - Camera device 2; overlay landmarks; flip; generate a recording

python intercept.py -i "/Downloads/shakira.mp4" --second 0 - Use video file as input; use camera device 0 as secondary input for mirroring feedback