Stereo Matching Based on the IC-Stereo Network in Low-Light Environments

In low-light environments such as nighttime scenes, reduced illumination often leads to the loss of image details and geometric information, making semantic understanding more challenging. Consequently, stereo matching performance is significantly degraded. To address this issue, this paper proposes a deep learning-based stereo matching network, IC-Stereo, which integrates contextual and geometric information to facilitate more effective cost aggregation, thereby improving disparity estimation in low-light conditions. Additionally, the network incorporates a low-light image enhancement module to restore image details, further enhancing the quality of the generated disparity maps. To evaluate the performance of the proposed model under low-light conditions, a synthetic low-light dataset was generated from the KITTI dataset using image processing techniques, and a real-world low-light dataset was captured using a binocular camera with low exposure settings. These datasets were used to assess the model's robustness in challenging lighting scenarios. Experimental results demonstrate that the integration of the context and geometry fusion module, along with the image enhancement module, significantly improves the model's ability to produce disparity maps with richer detail and smoother global structure, enabling high-accuracy stereo matching under low-light conditions. Moreover, to ensure the model retains strong performance under normal lighting, it was also evaluated on several public benchmark datasets. The results confirm that the proposed model not only surpasses traditional algorithms but also outperforms most state-of-the-art deep learning-based methods in terms of accuracy.