煤矿井下普遍存在低照度、弱纹理和结构化的特征退化场景，导致视觉SLAM(visual simultaneous localization and mapping)系统面临有效特征不足或误匹配率高的问题，严重制约了其定位的准确性和鲁棒性。

提出一种基于边缘感知增强的视觉SLAM方法。首先，构建了边缘感知约束的低光图像增强模块。通过自适应尺度的梯度域引导滤波器优化Retinex算法，以获得纹理清晰光照均匀的图像，从而显著提升了在低光照和不均匀光照条件下特征提取性能。其次，在视觉里程计中构建了边缘感知增强的特征提取和匹配模块，通过点线特征融合策略有效增强了弱纹理和结构化场景中特征的可检测性和匹配准确性。具体使用边缘绘制线特征提取算法(edge drawing lines, EDLines)提取线特征，定向FAST和旋转BRIEF点特征提取算法(oriented fast and rotated brief, ORB)提取点特征，并利用基于网格运动统计(grid-based motion statistics, GMS)和比值测试匹配算法进行精确匹配。最后，将该方法与ORB-SLAM2、ORB-SLAM3在TUM数据集和煤矿井下实景数据集上进行了全面实验验证，涵盖图像增强、特征匹配和定位等多个环节。

结果表明：(1) 在TUM数据集上的测试结果显示，所提方法与ORB-SLAM2相比，绝对轨迹误差、相对轨迹误差的均方根误差分别降低了4%~38.46%、8.62%~50%；与ORB-SLAM3相比，绝对轨迹误差、相对轨迹误差的均方根误差分别降低了0~61.68%、3.63%~47.05%。(2) 在煤矿井下实景实验中，所提方法的定位轨迹更接近于相机运动参考轨迹。(3) 有效提高了视觉SLAM在煤矿井下特征退化场景中的准确性和鲁棒性，为视觉SLAM技术在煤矿井下的应用提供了技术解决方案。研究面向井下特征退化场景的视觉SLAM方法，对于推动煤矿井下移动式装备机器人化具有重要意义。

Underground coal mines commonly exhibit low illuminance, weak textures, and structuralization-induced feature degradation. These scenes lead to challenges of insufficient effective features and high mismatch rates to the visual simultaneous localization and mapping (SLAM) system, severely compromising its localization accuracy and robustness.

This study proposed an edge awareness-enhanced visual SLAM method. First, an edge-awareness constrained low-illuminance image enhancement module was developed. Specifically, images with clear textures and uniform illumination were obtained using the Retinex algorithm optimized using an adaptive gradient-domain guided filter. This significantly improved feature extraction performance under low and uneven lighting conditions. Second, an edge awareness-enhanced feature extraction and matching module was introduced into the visual odometry. A point and line feature fusion strategy was employed to enhance the detectability and matching accuracy of weak textures and features in structured scenes. Specifically, line features were extracted using the EDLines algorithm, while point features were extracted using the Oriented FAST and Rotated BRIEF (ORB) algorithms. Such feature extraction was followed by precise feature matching achieved using grid-based motion statistics (GMS) and ratio test matching algorithms. Finally, the proposed method, along with the ORB-SLAM2 and ORB-SLAM3 algorithms, was comprehensively verified on the TUM dataset and the dataset of the actual underground coal mine scenes, covering multiple aspects such as image enhancement, feature matching, and localization.

The results indicate that on the TUM dataset, the proposed method reduced the root mean square errors (RMSEs) of absolute and relative trajectory errors by 4%‒38.46% and 8.62%‒50%, respectively compared to ORB-SLAM2 and reduced by 0‒61.68% and 3.63%‒47.05%, respectively compared to ORB-SLAM3. Experiments on the actual underground coal mine scenes revealed that the location trajectories of the proposed method were aligned with the reference trajectory of camera motion more closely. Furthermore, the proposed method effectively enhanced the accuracy and robustness of the visual SLAM system in the feature degradation scene in underground coal mines, providing a technical solution for its applications in such settings. Research on visual SLAM methods tailored for feature degradation scenes in underground coal mines holds great significance for advancing the roboticization of mobile equipment used in coal mines.