Abstract: Blast hole inspection robots in open-pit mining play a vital role in ensuring the safety of blasting operations and advancing intelligent mining. Traditional manual inspection methods suffer from low efficiency, limited accuracy, and high operational risk, making them inadequate for large-scale and efficient open-pit mining. To address these challenges, a binocular vision-based three-dimensional localization method for blast holes is proposed to provide high accuracy, real-time performance, and robustness for inspection robots. In the method design, a weighted Laplacian equation filling strategy guided by reference images is introduced to overcome the sparsity and severe noise of depth labels during training. This strategy achieves globally consistent completion of invalid depth regions, thereby improving the integrity and reliability of training data. A target detection network is then used to enable rapid detection and identification of blast hole regions, ensuring efficiency and accuracy in the subsequent stereo matching process. During stereo matching, an ECANet channel attention mechanism is incorporated into the iterative geometry encoding volume (IGEV−Stereo) network to enhance feature extraction capability. In addition, an edge-aware loss function based on Canny edge masks is designed to constrain boundary feature matching, effectively reducing matching errors at edges without increasing inference overhead. Finally, precise three-dimensional localization of blast holes is achieved using camera calibration parameters. Experimental results show that, on our self-constructed blast hole dataset, the target detection algorithm achieves an accuracy of 93.9%, and the improved stereo matching algorithm reduces the average end-point error (EPE) by 20.36% compared with the baseline. The single-frame inference latency is 1.8 ms for target detection and 77.44 ms for stereo matching, yielding an end-to-end latency of 79.24 ms, meeting real-time application requirements. In field tests conducted in real open-pit mining scenarios, the 3D depth error of blast holes reached up to 25 mm, highlighting high accuracy and strong robustness under complex blasting conditions. These findings provide new insights and effective technical support for the intelligent development of blast hole inspection robots and offer significant engineering application value for improving the safety and automation of mining operations.