
Rock Paper Scissors AI
Real-Time Gesture Recognition - No Buttons, Just Your Hands
“Rock, Paper, Scissors is a solved problem. Making it controller-free is the interesting part.”
Overview
Rock Paper Scissors AI is a hands-free implementation of the classic game, played in real time using only a webcam. The system uses MediaPipe's hand landmark model to extract the 3D positions of 21 hand keypoints per frame, then classifies the landmark configuration as rock, paper, or scissors with a custom classifier.
Gesture Classification
Each gesture is defined by the angular relationships between finger joints. A clenched fist (rock) has all finger joints folded. Paper has all fingers extended. Scissors has the index and middle fingers extended while the others are folded. The classifier computes joint angles from the x/y/z landmark coordinates, compares them against threshold ranges for each class, and produces a gesture prediction at real-time frame rates.
The Game Loop
The game uses a countdown to synchronise the player's gesture with the AI's randomly selected move, mimicking the real game's "rock, paper, scissors, shoot" cadence. Results are displayed as an overlay on the live camera feed, with a running score tracked across rounds. The entire experience runs locally: no server, no latency, just a camera and Python.
Project Gallery
7 items




Explore More Projects
Swipe or click arrows to see other work
















