Radar Pedestrian Classification via Spectral Feature Visualization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional radar target classification methods rely on neural networks, which are 'black boxes' and lack transparency, making it difficult to understand why decisions are made, and often require extensive training and learning algorithms, leading to unreliable and uninterpretable results.

Innovation Solution

An apparatus and method for automatic radar target classification using spectral data from radar signals, featuring a visualization module and a classification module that extract specific features to identify pedestrians and other objects, providing a transparent and reliable classification process by defining pre-determined tests based on unique characteristics of target types.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If neural networks are used for radar target classification, then classification capability is achieved, but transparency and interpretability of decisions are lost

Engineering Contradiction:
Improveclassification reliabilityVSAvoiddecision transparency
Core Design Contradiction:
ReliabilityVSLoss of information

Solution Approach 1:

The patent replaces the neural network 'black box' system with a rule-based classification system that uses explicit spectral feature analysis. Instead of relying on learned patterns from training data, the system directly evaluates spectral characteristics (peak positions, amplitudes, bandwidths) against predefined criteria to classify targets, making the decision process transparent and interpretable while maintaining classification reliability

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms the classification approach by changing from parameter learning (neural networks) to parameter measurement (spectral analysis). The system measures specific spectral parameters (frequency peaks, amplitudes, bandwidths) and uses these direct measurements for classification, eliminating the need for training algorithms while preserving decision transparency

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If neural networks with training algorithms are used, then classification accuracy may improve, but system complexity and computational requirements increase

Engineering Contradiction:
Improveclassification accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts and uses only the essential spectral features needed for classification (peak frequencies, amplitudes, bandwidths) without requiring the complex training and learning infrastructure of neural networks. By taking out only the necessary measurement and evaluation components, the system achieves classification accuracy with significantly reduced complexity

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent replaces expensive, complex neural network training systems with simple, direct spectral analysis methods. The classification is achieved through straightforward measurement and comparison of spectral parameters against predefined criteria, eliminating the need for computationally intensive training algorithms while maintaining effectiveness

Inventive Principle:
Principle #27Cheap short-living objects (Disposable)

3Reliability

If extensive training algorithms are applied, then model performance may improve, but processing time and computational resources increase

Engineering Contradiction:
Improveclassification reliabilityVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary action by pre-defining the classification criteria based on spectral characteristics before actual classification occurs. Instead of training during operation, the system has predetermined the decision rules for identifying pedestrians, vehicles, drones, and birds, enabling immediate classification without time-consuming training phases

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent skips the extensive training and learning phase entirely by using direct spectral analysis. The system rushes through to immediate classification by evaluating spectral features against predefined criteria, eliminating the time loss associated with neural network training while maintaining classification reliability

Inventive Principle:
Principle #21Skipping (Rushing through)

Data Source

PatentEP4030187A1Apparatus and method for identifying pedestrians in radar signals
Publication Date: 2022.07.20 HENSOLDT SENSORS GMBH
  • EP4030187A1 patent drawingFigure 1
  • EP4030187A1 patent drawingFigure 2~3B
  • EP4030187A1 patent drawingFigure 4A~4B

AI summary

An apparatus for automatic identification of a pedestrian in radar signals based on features of spectral data of the radar signals is disclosed. The apparatus comprises: a visualization module (no) configured to visualize feature-specific structures in the spectrum (spe) and a classification module (120), configured to identify the pedestrian based on predefined tests testing a presence of at least one feature and being performed on a result of visualization module (110).