A Learning-Based Method for Open-Pit Mine Dense Reconstruction Based on Vehicle-Mounted Camera

Environmental monitoring in open-pit mining plays a crucial role in ensuring not only improved productivity but also worker safety and environmental protection. The traditional monitoring techniques, such as GPS and electronic total stations used in open-pit mining, encounter challenges in data modeling and preprocessing. These difficulties require skilled personnel, resulting in high labor costs. To address the issues mentioned above, this paper proposes a novel learning-based dense reconstruction method for open-pit mines, which utilizes images captured by vehicle-mounted cameras. This innovative method enhances the accuracy of traditional 3D reconstruction techniques by integrating a residual network and feature pyramid. The proposed method achieved an accuracy score of 0.541, a completeness score of 0.325, and an average F1 score of 48.31. The experiment shows that our model outperforms the traditional Multi-View Stereo (MVS) method and captures the local details of the open-pit mine with high efficiency and low cost.