Efficient and accurate registration with BWPH descriptor for low-quality point clouds.

Area of Science:

• Computer Vision
• 3D Geometry Processing
• Robotics

Background:

• Point cloud registration is vital for 3D computer vision.
• Existing local descriptors face challenges with accuracy, memory, and speed, particularly for low-cost sensor data.
• These limitations hinder real-world applications requiring precise 3D reconstruction and analysis.

Purpose of the Study:

• To develop an efficient and accurate local descriptor for 3D point cloud registration.
• To address the limitations of current methods when processing data from low-cost sensors.
• To improve the speed and reduce the memory footprint of point cloud registration.

Main Methods:

• Proposed Binary Weighted Projection-point Height (BWPH) descriptor.
• Integration of Gaussian kernel density estimation, weighted height characteristics, and binarization.
• Extensive experimental validation and comparison with state-of-the-art techniques.

Main Results:

• BWPH descriptor demonstrates high matching accuracy and strong compactness.
• The method shows feasibility across various contexts and datasets.
• Successful registration of real-world datasets from low-cost sensors with minimal errors.

Conclusions:

• BWPH offers an efficient and accurate solution for 3D point cloud registration.
• The descriptor is particularly effective for data acquired by low-cost sensors.
• BWPH enables precise initial alignment positions, enhancing 3D vision applications.