SVM Traffic Sign Classification Database Generation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional methods for classifying traffic signs using Support Vector Machines (SVMs) face challenges in reducing computation time while maintaining performance, often resulting in high false positive detections and requiring frequent updates in digital map data.

Innovation Solution

A method is developed to generate a classification data base by retrieving the parameter set of a trained SVM, generating a symmetric or hermitian matrix, and determining eigenvectors, which reduces the number of vectors needed for classification, allowing for faster processing without compromising accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional SVM methods use a large number of support vectors to achieve acceptable classification accuracy, then classification reliability is improved, but detection time and training time become excessively large

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

Solution Approach 1:

The patent extracts only the most critical support vectors from the complete SVM model. By identifying and retaining a minimal subset of support vectors that contribute most significantly to classification decisions, the system achieves acceptable accuracy with dramatically reduced computational overhead during detection phases.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies different quality levels to different support vectors by assigning weights based on their individual contribution to classification. Rather than treating all support vectors equally, the system identifies high-value vectors that provide disproportionate benefit to classification accuracy, allowing selective retention of only these critical vectors.

Inventive Principle:
Principle #3Local quality

2Productivity

If conventional SVM methods reduce the number of support vectors to decrease detection time, then productivity is improved, but classification reliability deteriorates with unacceptable fail rates

Engineering Contradiction:
Improvedetection speedVSAvoidclassification accuracy
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent changes the parameter of support vector selection from a fixed large number to a dynamically determined optimal subset size. By analyzing the contribution of each support vector and selecting only those above a certain threshold, the system achieves the right balance between speed and accuracy rather than simply reducing vector count arbitrarily.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If conventional SVM methods use a large number of support vectors to maintain low error rates, then measurement precision is improved, but device complexity increases

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

Solution Approach 1:

The patent extracts the essential classification capability from a complex SVM model by removing redundant support vectors. The resulting simplified model retains the core decision-making functionality while eliminating unnecessary computational complexity, making the system more efficient without sacrificing essential accuracy.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP2442258B1Method of generating a classification data base for classifying a traffic sign and device and method for classifying a traffic sign
Publication Date: 2018.12.05 HARMAN BECKER AUTOMOTIVE SYST GMBH
  • EP2442258B1 patent drawingFigure 1~2
  • EP2442258B1 patent drawingFigure 3
  • EP2442258B1 patent drawingFigure 4

AI summary

Method of Generating a Classification Data Base for Classifying a Traffic Sign and Device and Method for Classifying a Traffic Sign A method of generating a classification data base (7) for traffic sign classification is described. Based on a parameter set (3) of a trained support vector machine, a symmetric matrix is generated. A set of mutually orthogonal eigenvectors of the matrix is determined. At least a sub set of the mutually orthogonal vectors is stored in the classification data base (7). The classification data base (7) may be deployed to devices (13,14) for classifying a traffic sign.