Multiview Image Composition With Feedback-Based Stitching Correction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional automotive visualization systems face challenges in creating accurate panoramic images due to camera limitations, leading to distorted views and artifacts, which can compromise driver safety by providing incomplete or inaccurate information.

Innovation Solution

A closed-loop composition quality optimization module is used to assess and optimize image composition from multiple cameras, applying geometric and photometric transforms to ensure proper alignment and color matching, thereby minimizing distortions and artifacts.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional homography estimation is used for image composition, then the composition process is simple and fast, but geometric distortion and misalignment artifacts occur in the stitched images

Engineering Contradiction:
Improvecomposition speedVSAvoidimage alignment accuracy
Core Design Contradiction:
ProductivityVSManufacturing precision

Solution Approach 1:

The patent implements a feedback mechanism where composition quality scores are computed for overlapping regions between adjacent images. Based on these quality scores, the system iteratively adjusts geometric transforms to maximize composition quality, ensuring proper alignment while maintaining processing efficiency through targeted optimization rather than exhaustive search.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The system dynamically adjusts geometric transform parameters (rotation, translation, scaling) based on computed composition quality scores. By changing these parameters iteratively to optimize alignment in overlapping regions, the system resolves geometric distortion while maintaining reasonable processing speed through efficient parameter search strategies.

Inventive Principle:
Principle #35Parameter changes

2Area of stationary object

If wider fields of view are attempted to capture more safety-critical areas, then more of the environment is visible, but image distortion and artifacts increase due to camera limitations

Engineering Contradiction:
Improvefield of view coverageVSAvoidimage quality
Core Design Contradiction:
Area of stationary objectVSManufacturing precision

Solution Approach 1:

The patent divides the wide field of view into multiple overlapping image regions captured by individual cameras. Each region is processed and composed with its neighbors using quality-based geometric transforms, allowing the system to maintain high local image quality while achieving comprehensive wide-area coverage through the combination of multiple segmented views.

Inventive Principle:
Principle #1Segmentation

3Measurement precision

If feature points are used for homography estimation, then correspondence between images can be established, but the method becomes ineffective for scenes without distinct objects and produces distortion in scenes with depth

Engineering Contradiction:
Improvefeature point correspondenceVSAvoidscene type applicability
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent replaces the mechanical feature-point-matching approach with a quality-score-based geometric optimization method. Instead of relying on distinct visual features, the system uses overlap region quality assessment to guide transform parameter adjustment, making the composition method effective for all scene types including those without distinct objects or with significant depth variations.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS11948315B2Image composition in multiview automotive and robotics systems
Publication Date: 2024.04.02 NVIDIA CORP
  • US11948315B2 patent drawing
  • US11948315B2 patent drawing
  • US11948315B2 patent drawing

AI summary

In various examples, two or more cameras in an automotive surround view system generate two or more input images to be stitched, or combined, into a single stitched image. In an embodiment, to improve the quality of a stitched image, a feedback module calculates two or more scores representing errors between the stitched image and one or more input images. If a computed score indicates structural errors in the stitched image, the feedback module calculates and applies one or more geometric transforms to apply to the one or more input images. If a computed score indicates color errors in the stitched image, the feedback module calculates and applies one or more photometric transforms to apply to the one or more input images.