Vehicle Vision System Geo-Referenced Feature Matching

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current vehicle vision systems struggle to reliably identify specific Points of Interest (POIs) such as buildings or landmarks, lacking robustness and speed in object recognition, especially under varying environmental conditions like lighting and viewing directions.

Innovation Solution

A vehicle vision system that employs a camera and processor to capture images, determine feature descriptors, and match them with geo-referenced feature descriptors from annotated map data, enabling robust and fast object recognition by utilizing geo-referenced feature vectors and position information for accurate orientation and positioning.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional object recognition methods are used, then classification into limited classes is reliable, but identification of specific POIs such as buildings or landmarks is not possible

Engineering Contradiction:
Improveobject recognition reliabilityVSAvoidspecific POI identification capability
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The system segments the object recognition task into two parts: using traditional classification methods for general object identification and using feature descriptor matching with annotated map data for specific POI identification. This allows the system to maintain reliability for common objects while extending capability to identify specific landmarks and buildings.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system uses pre-computed feature descriptors and annotated map data that are prepared in advance. By storing geo-referenced feature descriptors in the map data before the actual recognition task, the system enables fast and accurate POI identification without requiring real-time training or complex processing during execution.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If neural networks are trained for specific POI recognition, then identification accuracy improves, but system complexity and training requirements increase

Engineering Contradiction:
ImprovePOI identification accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

Instead of training a complex neural network from scratch for specific POI recognition, the system uses copies of pre-computed feature descriptors that are extracted from images and matched against annotated map data. This approach achieves high POI identification accuracy without requiring complex training procedures or large computational resources.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The system uses simple feature descriptor matching rather than complex neural networks. The feature descriptors act as lightweight, disposable representations that can be quickly extracted and matched, providing sufficient accuracy for POI identification without the overhead of trained neural network models.

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

3Reliability

If feature descriptors are used for robust recognition, then robustness against varying lighting and viewing directions improves, but processing speed decreases

Engineering Contradiction:
Improverecognition robustnessVSAvoidprocessing speed
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The system pre-computes and stores feature descriptors in the annotated map data before the actual recognition task. By having these descriptors ready in advance, the system can quickly match them against extracted features from images, achieving both robustness under varying conditions and fast processing speeds without requiring complex real-time computation.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP2450667B1Vision system and method of analyzing an image
Publication Date: 2016.11.02 HARMAN BECKER AUTOMOTIVE SYST GMBH
  • EP2450667B1 patent drawingFigure 1
  • EP2450667B1 patent drawingFigure 2
  • EP2450667B1 patent drawingFigure 3

AI summary

A vision system (1) comprises a camera (5) configured to capture an image (15) and a processor (2) coupled to the camera (5) and configured to determine at least one feature descriptor for the image (15). The processor (2) has an interface (3) to access annotated map data which includes geo-referenced feature descriptors. The processor (2) is configured to perform a matching procedure between the at least one feature descriptor determined for the at least one image (15) and the retrieved geo-referenced feature descriptors.