Grid-Generated CNN Parameter Mapping for Varied Road Layouts

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Convolutional Neural Networks (CNN) used in autonomous driving struggle with inefficiency when input images deviate from the typical arrangements used in training images, as parameters optimized for central roads may not properly process test images with differently arranged road features.

Innovation Solution

A method utilizing a grid generator to divide test images into subsections based on dynamic templates, incorporating non-object location and class information to determine optimized neural network parameters for testing, allowing for effective processing of varied road arrangements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If CNN parameters are optimized for training images with typical road arrangements (roads in centers), then processing efficiency is improved for those specific arrangements, but processing accuracy deteriorates when test images have different road arrangements

Engineering Contradiction:
Improveprocessing efficiencyVSAvoidprocessing accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent divides the input image into multiple regions of interest (ROIs) based on detected non-object locations. Each ROI is then processed independently with appropriate CNN parameters selected based on the regional characteristics. This segmentation allows different parameter sets to be applied to different spatial locations, resolving the contradiction between optimized processing for typical arrangements and accurate processing for varied arrangements.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent dynamically selects CNN parameters based on the actual content and arrangement of objects in each region of the input image. Rather than using fixed parameters optimized for a single typical arrangement, the system adapts parameter selection to match the specific spatial configuration found in each test image, thereby maintaining both efficiency and accuracy across diverse scenarios.

Inventive Principle:
Principle #15Dynamics

2Device complexity

If CNN uses fixed parameters optimized for typical training image arrangements, then model simplicity is maintained, but adaptability to different road arrangements deteriorates

Engineering Contradiction:
Improvemodel simplicityVSAvoidadaptability to different arrangements
Core Design Contradiction:
Device complexityVSAdaptability or versatility

Solution Approach 1:

The patent performs preliminary detection of non-objects and division into regions of interest before applying CNN processing. This preliminary action identifies the spatial arrangement characteristics of the input image, enabling subsequent dynamic parameter selection. The preprocessing step adds minimal complexity while significantly improving adaptability to different road arrangements.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent applies different CNN parameters to different regions of the input image based on local characteristics. Each region receives parameter optimization tailored to its specific content and arrangement, rather than applying a single global parameter set. This local adaptation maintains model simplicity overall while achieving high adaptability to various road configurations.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3686784B1Method and device of neural network operations using a grid generator for converting modes according to classes of areas to satisfy level 4 of autonomous vehicles
Publication Date: 2024.07.31 STRADVISION
  • EP3686784B1 patent drawingFigure 1
  • EP3686784B1 patent drawingFigure 2
  • EP3686784B1 patent drawingFigure 3

AI summary

A method of neural network operations by using a grid generator is provided for converting modes according to classes of areas to satisfy level 4 of autonomous vehicles. The method includes steps of: (a) a computing device, if a test image is acquired, instructing a non-object detector to acquire non-object location information for testing and class information of the non-objects for testing by detecting the non-objects for testing on the test image; (b) the computing device instructing the grid generator to generate section information by referring to the non-object location information for testing; (c) the computing device instructing a neural network to determine parameters for testing; (d) the computing device instructing the neural network to apply the neural network operations to the test image by using each of the parameters for testing, to thereby generate one or more neural network outputs.