American Sign Language (ASL) recognition is required to improving communication for people who are deaf or hard of hearing. Previous approaches based on machine learning models have significant limitations, such as variable accuracy and sensitivity to gesture variations. We aim to overcome these challenges by proposing an innovative approach using voting methods, in particular hard voting, to improve the accuracy of ASL gesture recognition. We have applied this method to the Kaggle ASL dataset. Our test results show that our approach significantly improves the accuracy of ASL gesture recognition, achieving a significantly higher accuracy rate than traditional systems, which have an accuracy of 60%. This improvement is demonstrated by the rigorous comparative tests carried out. In addition, we have developed a graphical user interface (GUI) that allows users to capture signs in real-time or to browse a sign image from a folder to obtain the appropriate alphabet. Our work thus aims to bridge the gap between the needs of ASL users and the capabilities of current recognition systems, offering a more accurate and user-friendly solution.