Adaptive Point Cloud Smoothing for Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
3D laser scanner point cloud files are distorted by noise such as statistical noise and false scatter points, leading to inaccurate representation of object dimensions, especially for precise measurements required in engineering applications.
Innovation Solution
A method for processing point cloud data by calculating local error limits for each pixel, adjusting distance values based on differences with neighboring pixels, and applying weighted averaging and repetitive averaging techniques to reduce noise and scatter points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If smoothing techniques are applied to 3D point cloud data, then measurement accuracy is improved, but the density of the original scan data is distorted and/or diminished
Solution Approach 1:
The patent applies different processing treatments to different regions of the point cloud data based on local characteristics. High-density regions are preserved with minimal smoothing, while low-density regions receive adaptive smoothing. This local differentiation allows measurement accuracy to be improved in regions where it matters without globally distorting the data density.
Solution Approach 2:
The smoothing algorithm dynamically adjusts its intensity and parameters based on the local density and characteristics of the point cloud data. The processing is not static but adapts to the specific conditions of each region, allowing the system to maintain data density where points are abundant while applying necessary smoothing where points are sparse, thus resolving the contradiction between measurement accuracy and data density preservation.
2Reliability
If aggressive noise removal is applied to eliminate statistical noise and scatter points, then measurement reliability is improved, but the true shape details of the object are lost
Solution Approach 1:
The patent employs multiple processing passes with varying parameters and thresholds. Different regions of the point cloud are processed with different noise removal aggressiveness levels based on local density, curvature, and other geometric parameters. This allows the system to achieve high measurement reliability in suitable regions while preserving fine shape details in regions where aggressive processing would cause loss of information.
Solution Approach 2:
Instead of applying uniform noise removal across the entire point cloud, the patent applies partial processing selectively to regions where it is most beneficial. The algorithm identifies and processes only the portions of the data that require noise removal, leaving other regions untouched or minimally processed. This partial action approach maintains measurement reliability where needed while preserving shape accuracy where aggressive processing would be harmful.
3Measurement precision
If registration markers are used to aid accurate measurement, then measurement consistency is improved, but the point cloud becomes distorted and does not accurately represent the object
Solution Approach 1:
The patent detects and extracts registration markers from the point cloud data, processes them separately to establish coordinate transformations, and then removes their distorting influence from the main object model. By separating the marker processing from the object processing, the system can achieve measurement consistency through proper registration while preventing the markers from distorting the representation of the actual object geometry.
Data Source
AI summary
A method is explained for processing an array of pixels in a point cloud. Local error bars are calculated for each pixel distance value. A difference is calculated between distance values of the pixel being processed and neighboring pixels with distance values within the error bars. If the difference is outside the error bars, the distance value of the pixel being processed is changed by a small fraction while remaining inside the error bars; if the difference is within the error bars, the pixel value is replaced by a weighted average. The neighboring pixels with distance values within the error bars of the pixel are counted and if a predetermined threshold is met, the counted values are averaged and replace the pixel value, but if not met, the pixel value is unchanged. If loop exit criteria have been met, the loop is terminated and if not, looping begins again.