【摘要】 提出了一种基于UNet图像分割与RAFT-Stereo立体匹配的双目视觉方法，用于高精度测量透明液体液位。针对传统双目立体匹配在透明液体场景中因纹理缺失导致的深度估计误差问题，该方法通过UNet网络对液面区域进行精确分割，生成像素级掩膜以突出液面特征；结合RAFT-Stereo算法计算视差并转换为深度信息，显著提升了液位测量的准确性与鲁棒性。实验结果表明，该方法显著优于传统半全局块匹配(SGBM)和RAFT-Stereo算法，平均绝对误差(MAE)和最大误差(MaxE)分别降低约85.2%和82.1%,有效改善了透明液体液面模糊和匹配失败问题。未来研究可进一步优化模型实时性，为航天燃料罐监测等实际应用提供可靠技术支持。更多还原

【Abstract】 A binocular vision method based on UNet image segmentation and RAFT-Stereo stereo matching for high-precision measurement of transparent liquid levels is proposed.To address the depth estimation errors caused by the lack of texture in transparent liquid scenarios, the method employs the UNet network to accurately segment the liquid surface, generating pixel-level masks to highlight liquid surface features.The RAFT-Stereo algorithm is then applied to compute disparity and convert it into depth information, thereby significantly improving the accuracy and robustness of liquid level measurement.Experimental results show that the proposed method outperforms the traditional semi-global block matching(SGBM)and RAFT-Stereo algorithms, reducing the mean absolute error(MAE)and maximum error(MaxE)by approximately 85.2 % and 82.1 %,respectively, and effectively mitigating issues such as blurry liquid surfaces and matching failures.Future research will further focus on optimizing the model’s real-time performance to provide reliable technical support for practical applications such as aerospace fuel tank monitoring.更多还原