A High‑Precision Road Network Construction Method Based on Deep Learning for Unmanned Vehicle in Open Pit

To solve the problem of time-consuming and low precision in updating the open-pit vehicle transportation network, a high precision road network model construction method for unmanned vehicles in open-pit mines is proposed. This method can be divided into two steps. In the first step, an improved deep learning image processing model named DeepLabv3 + C (DeepLabv3 + Concat) is presented. Then, the road information extracted by the DeepLabv3 + C network is used to construct a three-dimensional model of the open-pit mine road network. In the second step, aiming at the time-consuming problem of unmanned vehicle meeting in open-pit mines, a vehicle meeting strategy was proposed. This strategy is used to guide the navigation of unmanned vehicles in open-pit mines. Besides, the DeepLabv3 + C network is verified by comparing the mIOU (means Intersection Over Union), accuracy, and continuity of road image extraction with the mainstream networks. The road network model constructed in the first step is quantitatively analyzed, and its performance is compared with GPS trajectory clustering methods. At the end of the paper, vehicle running simulation is carried out on the road network model by using Unity (a 3D visualization simulation software). The results show that the road network model constructed by this method can meet the navigation requirements of unmanned vehicles in open-pit mines, and the feasibility of the vehicle meeting strategy is proved.