Point Cloud Confidence Filtering for Accurate Map Merging

The patent applies local quality by evaluating and filtering point cloud data at the individual point level using spatial confidence metrics. Each point is assessed independently for quality characteristics such as spatial distribution, intensity values, and geometric consistency, allowing high-quality points to be merged while rejecting low-quality points. This selective filtering approach maintains spatial accuracy in the merged map while still expanding coverage through inclusion of valid points from new scan frames.

The patent employs parameter changes by introducing and evaluating multiple spatial confidence metrics that quantify the quality of point cloud data. These metrics include spatial distribution density, intensity value consistency, geometric feature validity, and deviation from expected spatial patterns. By dynamically calculating these parameters for each point and comparing them against threshold values, the system reliably distinguishes between high-confidence and low-confidence data, ensuring only reliable points are incorporated into the map while maintaining overall data completeness.

The patent applies partial action by implementing a multi-threshold filtering strategy rather than a single binary threshold. Multiple confidence thresholds are established (e.g., high, medium, low confidence levels), allowing the system to selectively include points based on their quality. Points above the high threshold are definitely included, points below the low threshold are definitely rejected, and points in the intermediate range are evaluated using additional criteria or included with reduced weight. This partial filtering approach maintains spatial accuracy by rejecting clearly low-quality points while minimizing data loss by preserving potentially valid points in the intermediate confidence range.