Machine Learning Image Processing Algorithm for Driver Assistance

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing driver assistance systems face challenges in image quality, particularly under inadequate illumination, leading to reduced performance and increased noise and motion blur, which are not effectively addressed by current methods that require complex learning processes and manual training data sets, limiting their applicability across multiple driver assistance functions.

Innovation Solution

A method utilizing machine learning to improve raw image data quality by training a processing algorithm with paired input and reference images from sensors with different qualities, allowing automatic parameterization and compensation for quality deficiencies, such as underexposure, through a convolutional neural network that can be applied across various driver assistance systems.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Illumination intensity

If exposure time is increased to improve image quality in low light conditions, then image brightness is improved, but image noise and motion blur increase

Engineering Contradiction:
Improveimage brightnessVSAvoidimage noise and motion blur
Core Design Contradiction:
Illumination intensityVSReliability

Solution Approach 1:

A machine learning-based processing algorithm acts as an intermediary between the raw sensor data and the final image output. The algorithm learns optimal processing parameters from training data to improve image quality without requiring increased exposure time, thus avoiding the trade-off between brightness and noise/blur

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The system changes processing parameters (such as denoising strength, sharpness, contrast) dynamically based on learned patterns from training data. These parameter adjustments allow image quality improvement in low-light conditions without physically changing sensor exposure settings that would cause motion blur

Inventive Principle:
Principle #35Parameter changes

2Manufacturing precision

If digital post-processing is applied to improve specific image quality features, then specific quality features are improved, but other quality features are degraded

Engineering Contradiction:
Improvespecific image quality featureVSAvoidother image quality features
Core Design Contradiction:
Manufacturing precisionVSReliability

Solution Approach 1:

The machine learning processing algorithm is designed to handle multiple image quality aspects simultaneously (noise reduction, sharpness, contrast, color accuracy) through a unified model. This multi-functional approach avoids the trade-offs of sequential single-purpose processing steps

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system dynamically adjusts multiple processing parameters simultaneously based on learned relationships from training data, optimizing overall image quality rather than maximizing individual features at the expense of others

Inventive Principle:
Principle #35Parameter changes

3Reliability

If separate training is performed for each driver assistance function, then function-specific performance is improved, but development effort and complexity increase

Engineering Contradiction:
Improvefunction-specific performanceVSAvoiddevelopment effort
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

A single machine learning processing algorithm is trained to improve image data for multiple driver assistance functions simultaneously. The unified model learns general image quality improvement patterns that can be applied across different functions (object detection, lane recognition, traffic sign recognition), reducing development complexity while maintaining function-specific performance

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The training processes and processing algorithms for multiple driver assistance functions are merged into a single unified machine learning model. This consolidation reduces the overall development effort and complexity while maintaining the ability to serve multiple functions

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP3291172B1Method for calculating final image data
Publication Date: 2020.03.11 AUDI AG
  • EP3291172B1 patent drawingFigure 1~2
  • EP3291172B1 patent drawingFigure 3~4

AI summary

Method for determining result image data, in particular in a motor vehicle (4), wherein raw image data are acquired by an imaging sensor (5), after which these are further processed by a processing algorithm for quality improvement in order to determine the result image data, wherein the processing algorithm is parameterized by several processing parameters which are determined by training the processing algorithm with several training data sets (1) within the framework of machine learning.