Multi-Sensor Point Cloud Consensus for Reflection Artifact Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional point cloud denoising techniques are ineffective in removing reflection noise or other types of artifacts from point cloud data, particularly in adverse weather conditions, leading to false object detections and inaccurate Lidar-based systems.
Innovation Solution
A method using multi-sensor consensus to filter-out artifacts by combining point clouds from multiple sensors, analyzing fused data with simple analytical geometry calculations to identify actual and expected sensor observations, and determining artifacts based on sensor counts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional point cloud denoising techniques are used to remove outliers from point clouds, then Gaussian noise can be filtered, but reflection noise and other artifacts with dense and large-scale noise cannot be effectively removed
Solution Approach 1:
The patent changes the filtering parameters and criteria by introducing multi-sensor consensus mechanisms and multiple filtering stages (initial filtering with statistical methods, followed by reflection-specific filtering using intensity analysis and geometric consistency checks). This allows the system to adapt to different noise types including Gaussian noise, reflection noise, and artifacts, resolving the contradiction between filtering accuracy and noise type coverage
Solution Approach 2:
The patent combines multiple filtering techniques into a composite filtering system that integrates statistical-based filtering, neighborhood-based filtering, and reflection-specific filtering methods. This composite approach enables the system to handle diverse noise types effectively, achieving both high reliability in artifact filtering and broad adaptability across different noise conditions
2Reliability
If intensity-based filtering is used to remove reflection points, then some reflection artifacts can be filtered, but true object points are incorrectly removed and not all erroneous points are eliminated
Solution Approach 1:
The patent introduces intermediate filtering stages and consensus mechanisms that act as mediators between raw intensity-based filtering and final object detection. Multiple sensors and filtering passes serve as intermediaries to verify point validity, preventing premature removal of true object points while eliminating reflection artifacts, thus resolving the contradiction between reflection filtering and object detection precision
Solution Approach 2:
The patent implements feedback mechanisms where filtering results are continuously evaluated and refined through multiple passes. The system uses feedback from multi-sensor data and geometric consistency checks to adjust filtering thresholds and criteria, preventing both over-filtering of true objects and under-filtering of reflections, thereby maintaining measurement precision while achieving reliable artifact removal
3Device complexity
If simple analytical geometry calculations are used in multi-sensor consensus, then computational complexity is reduced, but filtering effectiveness must be maintained
Solution Approach 1:
The patent segments the filtering process into distinct stages: initial data collection from multiple sensors, geometric consistency checking using simple analytical calculations, multi-sensor consensus verification, and final artifact removal decisions. This segmentation allows the use of computationally simple operations at each stage while maintaining overall filtering effectiveness through the cumulative power of the multi-stage process
Data Source
AI summary
A method for using multi-sensor consensus to filter-out artifacts from point cloud data includes receiving point cloud data from plurality of sensors, wherein point cloud data has at least two point clouds; generating fused point cloud by combining at least two point clouds; analyzing fused point cloud for identifying at least one object candidate present in real-world space; determining first set of sensors which actually observe at least one object candidate, based on point cloud data; estimating second set of sensors which are expected to observe at least one object candidate, based on geometry of at least two point clouds and pre-known poses of plurality of sensors; and determining whether given cluster including cluster points representing given object candidate is artifact, based on counts of sensors in first set and in second set of sensors.