Imagery Network Fine Registration by Reference Point Cloud Data Based on the Tie Points and Planes.

Area of Science:

• Geomatics Engineering
• Computer Vision
• Photogrammetry

Background:

• Cameras and laser scanners are crucial for 3D data acquisition but suffer from misalignment due to systematic and random errors.
• Accurate registration of multi-sensor data (imagery and point clouds) is essential for reliable 3D reconstruction and analysis.

Purpose of the Study:

• To propose a novel feature-based approach for fine registration of imagery and point cloud data.
• To enhance the accuracy and robustness of multi-sensor data alignment.

Main Methods:

• A differential tie plane is generated by matching tie points and their neighboring pixels in overlapping images.
• Preprocessing removes non-robust tie points; initial orientation parameters transform tie plane points to object space.
• Directional Vectors (DV) are calculated using nearest point cloud points, constraining tie points to the point cloud differential plane via collinearity equations.

Main Results:

• The proposed method achieved approximately 2.5 pixels error on checkpoints in both indoor and outdoor experiments.
• Demonstrated robustness and practicality in aligning image and point cloud data.

Conclusions:

• The novel feature-based approach effectively registers multi-sensor imagery and point cloud data.
• The method provides a robust and practical solution for accurate 3D data alignment, reducing errors significantly.