Simple and efficient registration of 3D point cloud and image data for an indoor mobile mapping system

Area of Science:

• Computer Vision
• Robotics
• Geospatial Data Science

Background:

• Accurate registration of 3D lidar point clouds and optical images is essential for multisource data fusion.
• Existing pose data often contains geometric misalignments between lidar and optical sensors.

Purpose of the Study:

• To develop a simple and efficient registration method to improve the accuracy of initial pose estimation.
• To enhance the applicability of integrating 3D point cloud data with optical imagery.

Main Methods:

• Feature extraction from single-frame lidar point clouds.
• Image feature extraction using the Canny edge detector.
• Construction of a cost map based on Canny edge detection to guide optimization.
• Utilization of a loss function to enhance method robustness.

Main Results:

• The proposed method effectively extracts features from both lidar and image data.
• A cost map guides the optimization process towards accurate registration.
• Experimental results demonstrate positive outcomes in aligning 3D lidar data with optical images.

Conclusions:

• The developed method offers a straightforward yet effective approach for 3D lidar and optical image registration.
• Improved pose accuracy facilitates better integration of multisource geospatial data.
• The technique shows promise for applications requiring precise sensor fusion.