A hybrid CNN-random forest model with landmark angles for real-time Arabic sign language recognition

Sign language serves as a dynamic means of communication, facilitating expression and emotional connection among individuals who are deaf or hard of hearing and the broader community. However, persistent communication barriers impact various sectors, including healthcare and education. Sign Language Recognition (SLR) systems bridge this gap by leveraging artificial intelligence advancements and linguistic knowledge to classify and translate sign language gestures in real-time. Despite advancements in SLR for many languages, Arabic SLR remains underexplored, lacking diverse datasets and real-time evaluation. This study introduces a novel Arabic SLR system emphasizing real-time applicability. The system achieves competitive performance in static and real-time classification through a hybrid CNN-Random Forest model trained on 96,000 collected images, incorporating hand landmarks’ angles. The proposed model demonstrates superior performance in real-time settings, surpassing existing alternatives and achieving an impressive 99.95% accuracy on aggregated data. These results highlight the model’s effectiveness in addressing the unique challenges of real-time Arabic SLR.