| 摘要 |
Systems and methods for merging three-dimensional models, such as a three-dimensional range sensor-based model and a three-dimensional camera-based model, are provided. According to aspects of the present disclosure, an enhanced volumetric merging technique can be used to merge the three-dimensional models. A plurality of voxels can be constructed for a three-dimensional space. A first distance field can be propagated based on the range sensor-based model in an extended margin between the range sensor-based model and a range sensor viewpoint. A second distance field can be propagated based on the camera-based model for voxels in the extended margin. A cumulative distance field can be determined based at least in part on the first field and the second distance field. The merged three-dimensional model can be constructed from the cumulative distance field using, for instance, a suitable meshing algorithm. |
| 主权项 |
1. A computer-implemented method for constructing a merged three-dimensional model, comprising:
determining, with a processing device, a first distance value for each of a plurality of voxels in an extended margin, the first distance value determined based on a range sensor-based model constructed from range sensor data, the extended margin extending between the range sensor-based model and a range sensor viewpoint associated with the range sensor-based model; determining, with the processing device, a second distance value for each of the plurality of voxels in the extended margin based upon a camera-based model constructed from imagery captured by a camera; determining, with the processing device, a cumulative distance value for each of the plurality of voxels in the extended margin, the cumulative distance value for a respective voxel in the plurality of voxels being determined based at least in part on the first distance value associated with the respective voxel and the second distance value associated with the respective voxel; and constructing, with the processing device, the merged three-dimensional model based at least in part on the cumulative distance value for each of the plurality of voxels. |
