Multi-Camera Soiling Detection Using Neural Image Classification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing systems for detecting lens soiling on vehicle-mounted image capture devices are inadequate, as they fail to effectively utilize multiple image sources and environmental conditions to enhance decision-making and accuracy in soiling detection.

Innovation Solution

A neural network image processing apparatus utilizing convolutional neural networks (CNNs) processes images from multiple image capturing devices with different fields of view, incorporating sensor data to enhance soiling detection and classification, allowing for end-to-end training and synergized decision-making.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a single image capturing device is used for soiling detection, then the device complexity is reduced, but the measurement precision of soiling detection deteriorates

Engineering Contradiction:
Improvesoiling detection accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent combines multiple image capturing devices with different fields of view into a unified soiling detection system. The neural network processor integrates images from multiple devices to comprehensively assess soiling conditions, thereby improving detection accuracy while managing system complexity through centralized processing.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces a new dimension of detection by using multiple image capturing devices with different fields of view (front, rear, side cameras) rather than relying on a single device. This multi-dimensional approach allows the system to detect soiling from various angles and perspectives, significantly enhancing detection precision.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Measurement precision

If multiple image capturing devices are used for soiling detection, then the measurement precision improves, but the device complexity increases

Engineering Contradiction:
Improvesoiling detection accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The neural network processor is designed with multi-functionality to handle various image processing tasks simultaneously - soiling detection, object detection, and semantic segmentation. This universal processor manages multiple image capturing devices and performs diverse functions, thereby improving detection precision without proportionally increasing overall system complexity.

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

Solution Approach 2:

The neural network processor acts as an intermediary that receives and integrates data from multiple image capturing devices. It processes and synthesizes information from different sources to produce accurate soiling detection results, mediating between the complexity of multiple sensors and the need for precise measurement.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If soiling detection is performed before object detection and semantic segmentation, then the reliability of vehicle systems is improved, but the loss of time in processing increases

Engineering Contradiction:
Improvevehicle system reliabilityVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system performs soiling detection as a preliminary action before object detection and semantic segmentation. By detecting soiling conditions first, the system can determine whether subsequent processing is necessary, thereby improving reliability while potentially reducing overall processing time by avoiding unnecessary computations when soiling is present.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The processing workflow is made dynamic based on soiling detection results. When soiling is detected, the system adjusts its processing sequence and priorities, dynamically allocating computational resources to maintain reliability while minimizing processing time losses.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3657379B1A neural network image processing apparatus for detecting soiling of an image capturing device
Publication Date: 2026.02.18 CONNAUGHT ELECTRONICS
  • EP3657379B1 patent drawingFigure 1
  • EP3657379B1 patent drawingFigure 2

AI summary

A neural network image processing apparatus obtains a first image and a respective additional image, wherein a first image capturing device has a field of view different than each additional image capturing device. Each image is processed through respective instances of a common feature extraction processing network to produce a respective first map and at least one additional map. Feature classification comprises processing the first map to provide a first classification indicating a soiling of the first image; processing the or each additional map to provide at least one additional classification indicating a respective soiling of the or each additional image; and combining the first and the or each additional classification to provide a boosted classification indicating soiling of the first image. If the boosted classification doesn't indicate soiling of the first image capturing device, further processing can be performed.