<p>The replacement of drill rods in coal mine drilling operations is a critical and challenging task, particularly in complex and confined environments. To address these challenges, this study presents an obstacle detection and path planning method that leverages depth cameras and adaptive bounding box technology. First, point cloud data of the drilling area are obtained using two depth cameras, and obstacles are identified using deep learning methods. To enhance the efficiency of path planning and address the issue of false collisions caused by the proximity of the drill rod installation position to obstacles, an adaptive radial depth octree partitioning (ARDOP) is developed. This method dynamically adjusts the fitting depth of octree-based bounding boxes according to the distance between points and the target position, effectively reducing the number of bounding boxes while ensuring high fitting accuracy near the target. In the field of path planning, this paper presents an improvement to the Bidirectional Rapidly exploring Random Tree (Bi-RRT) algorithm by incorporating a linear path cost function, thereby generating more optimal paths. The proposed methods were implemented on the Robot Operating System (ROS) platform and validated through simulation experiments. Experimental results demonstrate that the method accurately reconstructs drilling scenarios, and the adaptive bounding box algorithm reduces the number of bounding boxes by 55% compared to the octree bounding box with a depth of 7, enhancing path planning efficiency. The improved Bi-RRT algorithm (CBB-RRT) successfully avoids the generation of abnormal paths. This work provides a systematic solution for the intelligent operation of drill rod replacement in coal mines, showing significant potential for practical applications.</p>