Non-Uniform Voxelisation for Point Cloud Compression

Area of Science:

• Computer Vision
• 3D Data Processing
• Geometric Compression

Background:

• Point cloud compression is crucial for 3D data applications like VR and autonomous driving.
• Existing methods often use uniform voxelisation, which can struggle with complex scene geometries.
• This uniform approach may lead to loss of detail and less accurate reconstructions.

Purpose of the Study:

• To propose a novel non-uniform voxelisation technique for point cloud geometry compression.
• To improve the efficiency and accuracy of 3D data compression.
• To address the limitations of uniform voxelisation in capturing complex geometric details.

Main Methods:

• Developed a non-uniform voxelisation method that adaptively adjusts voxel sizes based on local point density.
• Implemented a technique to preserve geometric details during compression.
• Evaluated the method on benchmark datasets: ScanNet, ModelNet, and ShapeNet.

Main Results:

• The proposed non-uniform voxelisation achieved superior compression ratios compared to traditional uniform methods.
• Demonstrated enhanced reconstruction quality with the adaptive voxel sizing approach.
• Experimental results confirmed the effectiveness on diverse 3D datasets.

Conclusions:

• Non-uniform voxelisation is a promising alternative for point cloud geometry compression.
• The adaptive method offers improved performance for real-world 3D data scenarios.
• This technique enables more effective storage and transmission of complex 3D data.