Abstract: Road boundary detection is one of the key technologies for unmanned driving in mining areas. The obtained road boundary information can be used to assist the perception,planning and positioning of unmanned mining vehicles. Accurate road boundary detection and road boundary map construction are also the first step in the construction of high-precision maps. The road structure topology can be preliminarily constructed on the basis of existing road bounda- ry points through machine learning methods. Different from the traditional Kalman frame of road boundary tracking,this paper proposes a road boundary tracking scheme based on the idea of occupied grid. The so-called occupancy grid re- fers to a method for constructing scenes in the global scope based on obstacle information sensed by sensors. Due to the high uncertainty of road boundarys in mining areas,a model containing probability information is needed to find the lo- cation of the road boundary. The structure based on the octree can meet this demand. This paper screens the candidate road grids by rasterizing each frame of the real-time point cloud,and then performs a multi-feature road boundary de- tection based on ring compression. Among them,the grid is a point cloud structure that has been widely used. The coarse-grained nature of the mine road itself extends the feature detection to each fan-shaped cell as an inseparable basic unit,which improves the detection efficiency and avoids the interference caused by many road bumps. The road boundary tracking strategy uses a three-dimensional occupancy grid based on an octree,merges the multi-frame detec- tion results,and establishes a global road boundary map,extending the concept of the traditional two-dimensional occu- pancy grid map to a three-dimensional road boundary attribute global obstacle map. The actual mine road detection re- sults show that the proposed road edge detection algorithm for mine unstructured roads can accurately detect the mine road boundaries and can meet the requirements on the real-time driving of unmanned mine vehicles.