A hyper-optimized Deep Learning architecture capable of translating 10 static and dynamic hand gestures into text with 99.00% precision. Engineered using a 100% custom-curated dataset.
Unlike generic models relying on standard Kaggle datasets (like INCLUDE), this AI was trained on thousands of frames recorded entirely from scratch by the author. This guarantees zero domain-gap for the targeted gestures.
To achieve 99% accuracy in diverse environments, the dataset was artificially expanded by 500% using complex coordinate scaling, temporal shifting, and Gaussian noise injections.
OpenCV extracts 30 continuous frames per sequence.
MediaPipe isolates 225 spatial (X,Y,Z) coordinates.
A deep sequential neural network processes the temporal data.
Translates sequence into English text instantly.
Yes / Approval
No / Disapproval
Stop / Wait
Solid / Hold
Win / Peace
Perfect
Super / Rock On
Contact
Directional
Directional