Gesture Recognition translator using Machine Learning, Computer Vision, & MediaPipe

Abstract

Hand gesture recognition technology is transforming how humans interact with computers, especially in contactless interfaces and assistive communication. This paper presents the design and evaluation of a real-time hand gesture recognition system that combines Google's MediaPipe for landmark detection with classic machine learning on robust geometric features. The method strategically selects only critical, distance-based features between the wrist and fingertips and from the thumb to the other fingertips, resulting in a lightweight yet highly accurate classifier. Supporting five distinct static gestures—hello, good, yes, no, and thank you—the system achieves near-perfect recognition under typical webcam lighting. The paper details all stages, from dataset creation to live testing, and discusses plans for expanding gesture vocabulary and integrating audio feedback for multimodal interaction.