Radar Point Cloud Preprocessing for Low-Complexity Noise Filtering

Solution Overview

Problem

Existing noise removal algorithms for millimeter-wave radar point clouds are complex and require significant computing power, lacking a simple and effective solution tailored for this specific sensor type.

Innovation Solution

A preprocessing method involving reception intensity filtering, cluster analysis, and artificial intelligence model input to filter and identify target points in radar point clouds, utilizing top N1 and N2 points with the strongest reception intensity and cluster analysis to reduce noise.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If existing noise removal algorithms (Iteratively Least Square, neighborhood-based filtering, growing neural gas networks) are used for millimeter-wave radar point clouds, then noise filtering capability is improved, but algorithm complexity increases and significant computing power is required

Engineering Contradiction:

Improvenoise filtering capabilityVSAvoidalgorithm complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent extracts and applies LiDAR-specific noise removal algorithms (Iteratively Least Square, neighborhood-based filtering, growing neural gas networks) to millimeter-wave radar point clouds, adapting proven techniques from one domain to another while maintaining their effective noise filtering capabilities

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent makes noise removal algorithms universal by demonstrating their applicability across different sensor types (LiDAR and millimeter-wave radar), allowing the same algorithmic approaches to serve multiple sensing modalities and reducing the need for separate specialized solutions

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Reliability

If existing noise removal algorithms are used for millimeter-wave radar point clouds, then noise filtering capability is improved, but computing power requirements increase

Engineering Contradiction:

Improvenoise filtering capabilityVSAvoidcomputing power requirement

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent extracts and applies LiDAR-specific noise removal algorithms (Iteratively Least Square, neighborhood-based filtering, growing neural gas networks) to millimeter-wave radar point clouds, adapting proven techniques from one domain to another while maintaining their effective noise filtering capabilities

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent makes noise removal algorithms universal by demonstrating their applicability across different sensor types (LiDAR and millimeter-wave radar), allowing the same algorithmic approaches to serve multiple sensing modalities and reducing the need for separate specialized solutions

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12548278B2Preprocessing method and electronic device for radar point cloud

Publication Date: 2026.02.10 QUANTA COMPUTER INC

AI summary

A preprocessing method for a radar point cloud for object recognition is provided. A plurality of points from a radar are received. Each point indicates the reception intensity of the signal reflected from the object received by the radar. The points are filtered according to the reception intensity to obtain a plurality of first preprocessing points. A cluster analysis algorithm is executed on the first preprocessing points to identify a plurality of target points corresponding to the object in the first preprocessing points. The target points are filtered according to the reception intensity to obtain a plurality of second preprocessing points. The second preprocessing points are input into an artificial intelligence model to perform object recognition.