While deep learning techniques have demonstrated impressive outcomes in stereo matching of color images, it remains uncertain whether they can be effectively applied to stereo matching involving infrared images featuring projected textures. This study introduces an innovative approach for refining depth using deep learning, circumventing the necessity for collecting infrared image datasets and ensuring adaptability to various infrared camera models. Additionally, we evaluate the utility of infrared textures within the framework of deep learning and illustrate their continued efficacy in stereo matching despite the absence of color information. We introduce a fusion technique merging deep learning-based estimations with outcomes from traditional methods, enhancing the overall depth map’s comprehensiveness and precision.