Hand sign recognition systems play an increasingly important role in human–computer interaction and assistive communication technologies. Traditional unimodal approaches that rely solely on RGB images or geometric landmarks often struggle with variations in lighting, background, and signer appearance. In this paper, we propose a multimodal feature fusion framework that combines visual information from a pre-trained VGG16 network with structural hand landmark features processed through a custom 1D CNN. We introduce a novel 11-class hand sign dataset collected in controlled conditions and provide a comprehensive performance evaluation. Experimental results demonstrate that the proposed approach achieves 99.09% classification accuracy, outperforming RGB-only and landmark-only baselines by 2.29% and 9.59%, respectively. These findings confirm that integrating complementary modalities leads to more robust and discriminative representations. The approach stays efficient, making it suitable for real-time use in interactive systems.