An Occupancy Mapping Method Based on K-Nearest Neighbours.

Area of Science:

• Robotics
• Computer Vision
• 3D Mapping

Background:

• OctoMap is an efficient probabilistic framework for 3D occupancy mapping using octrees.
• Existing OctoMap update policies have limitations in handling noisy data and sensor occlusion.

Purpose of the Study:

• To develop an improved mapping method addressing OctoMap's limitations.
• To enhance occupancy probability estimation using contextual information.

Main Methods:

• A novel mapping method using the context of neighboring points to update occupancy nodes.
• Occupancy information is determined by the average distance to k-Nearest Neighbors (k-NN).
• A relationship between distance and probability change is defined using the Cumulative Density Function (CDF).

Main Results:

• The proposed method demonstrated improved occupancy probability estimation compared to standard OctoMap.
• Experiments on 20 datasets showed up to a 10% performance improvement over OctoMap's optimal settings.
• The method can decrease node probability even with points present, by leveraging neighbor distances.

Conclusions:

• The new method effectively overcomes OctoMap's limitations in probabilistic 3D mapping.
• Contextual information from k-NN significantly enhances occupancy estimation accuracy.
• This approach offers a more robust and accurate solution for 3D environment mapping from point clouds.