Parameter Reduction and Optimisation for Point Cloud and Occupancy Mapping Algorithms

Area of Science:

• Robotics
• Computer Vision
• 3D Environment Modeling

Background:

• Occupancy mapping generates 3D environment models from point clouds for robot navigation.
• Parameter selection significantly impacts mapping quality, but optimal configurations are under-researched.

Purpose of the Study:

• To develop a principled methodology for identifying and optimizing parameters for occupancy mapping.
• To enhance the performance and accuracy of 3D environment models.

Main Methods:

• A two-step approach: Neighbourhood Component Analysis for parameter significance and grid search with Area Under the ROC Curve for optimization.
• Utilized 20 datasets with precise ground truths for evaluation.
• Tested on OctoMap using point clouds from StereoSGBM stereo camera images.

Main Results:

• Identified mapping parameters as more critical than point cloud generation parameters.
• Achieved up to a 15% improvement in mapping performance compared to default parameters.
• Demonstrated the effectiveness of the proposed methodology in optimizing occupancy mapping.

Conclusions:

• The developed methodology effectively identifies and optimizes key parameters for occupancy mapping.
• Optimized parameters lead to significant improvements in 3D environment model quality.
• Highlights the importance of mapping parameters over point cloud generation parameters for performance.