Glaucoma, a leading cause of vision loss, often remains undiagnosed due to reliance on subjective assessments. This research leverages deep learning (DL) and computer vision (CV) to develop an accurate and objective glaucoma detection system. Our approach makes a hybrid model which combines convolutional neural networks (CNNs) through ensemble learning and a maximum voting-based strategy, achieving superior performance over traditional methods. The proposed system was rigorously evaluated on four public datasets, showing significant improvements over traditional methods. For binary classification, the model achieved 93.20% accuracy on the Rimone dataset and 99.29% accuracy on the Acrima dataset. For multimodal classification, it reached 95.53% accuracy on the Drishti dataset and 84.43% accuracy on the HVD dataset. These results highlight the model’s robust performance across various data types and conditions. In summary, this study advances medical imaging by providing a scalable, accurate, and interpretable system for early glaucoma detection. By addressing the limitations of traditional methods, our work holds the potential to significantly improve global patient outcomes through timely and reliable glaucoma diagnosis.