What is multi-view point cloud matching?
Substantially different angles to the object acquired point cloud data to match together, thereby establishing a complete and accurate three-dimensional model.
Most of the current multi-view matching method requires certain information to predict the point cloud data, affecting the algorithm versatility, autonomy.
1. For the scan order is unknown and can not extract any feature point cloud data presented based on criteria dual multi-angle of the initial acquisition method.
2. In response to the problems accumulated error of the initial three-dimensional model, a multi-angle precision registration method based on the moving average of improvement .
Multiview registration: the laser scanner can not be collected from a single perspective of the complete information of the object or scene, hence the need for a multi-view point cloud point cloud registration techniques, collected at different angles to match together, thereby generating a complete three-dimensional model.
Compared with the conventional image data, point cloud data can be more accurately and carefully to characterize the surface of the depth information of the geometry, and in addition without determining complex.
Multi-view point cloud registration is based on dual-view point cloud registration, that is the general point cloud matching.
Cloud points for matching two different angles, different times, different to the image forming apparatus acquired point cloud data, computing the transformation between them.
The cloud point is usually referred to as two data point cloud point cloud model:
Point cloud data calculated by the relationship between rotation and translation model point cloud, to find the optimal transformation between two point clouds.
When there is some overlap between the data and the model point cloud point cloud, by registration method dual view match them together.
For multi-angle, two adjacent stitching.
Emerging issues:
Accumulated error.
Matching the multi-view point cloud: eliminate the accumulated errors, an accurate three-dimensional model generated.
Practice, the point cloud data disorderly, how to solve? .
Research Status:
When a smaller number midpoint of the point cloud data, also known as point sets.
Iterative closest point algorithm: complete overlap registration problem between the point set. ICP
Disadvantage: when the local convergence case, the model can not be solved with the point set of data points is only partially overlap.
Crop iterative closest point: introducing a trimming ratio, effectively select a corresponding portion of focus points, thereby using the ICP algorithm to overlap part of registration,
Cons: local convergence
Iterative Closest Point:
Match point clouds objectives: to find the optimal rotation matrix and translation vector $ \ mathbf {R} \ in \ mathbb {R} ^ {3 \ times 3}, \ overrightarrow {t} in the space \ in \ mathbb {R} $ ^ {3}, so that after a rotation and translation transformation matched together two point clouds.