A Computer Vision-Based System for LESHO: Implementation of a Software System for the Translation of Static Gestures

Abstract

Language enables the transmission of information, ideas, and emotions; however, significant communication barriers persist for the deaf community in many everyday contexts. This paper presents a real-time software system for the translation of Honduran Sign Language (LESHO) based on computer vision and machine learning techniques. The proposed system relies on structured spatial representations extracted from anatomical landmarks of the hands and contextual body cues, rather than raw image data, enabling efficient and robust gesture recognition under real-time constraints. Gesture acquisition, feature extraction, and classification are integrated within a modular architecture implemented in Python using OpenCV, MediaPipe, and Scikit-learn. The system supports fixed-length landmark vectors for static gestures and temporal landmark sequences for dynamic gestures. Experimental evaluation on a multi-class dataset demonstrates high recognition accuracy and interactive response times on consumer-grade hardware. Overall, the results validate the effectiveness of landmark-based modeling for sign language recognition and highlight the system’s scalability for expanding vocabulary coverage and supporting inclusive communication in both face-to-face and technology-mediated environments.