Fisheye Camera 3D Projection for Undistorted Surround View

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Combining images from multiple fisheye cameras with different tilt angles and wide baselines poses challenges due to distortion and alignment issues, particularly in advanced driver assistance systems where a clear 360-degree surround view is required, as existing methods like panorama stitching fail to effectively handle large FOVs and distortions.

Innovation Solution

The system models each fisheye camera as a 3D shape, projects images onto a common 3D shape, and then aligns and stitches them onto a common plane to generate a combined view, compensating for tilt and rotation to minimize alignment errors and preserve image content.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Area of moving object

If wide-angle lenses are employed to capture large FOV images, then the field of view is improved, but image distortion increases

Engineering Contradiction:
Improvefield of viewVSAvoidimage distortion
Core Design Contradiction:
Area of moving objectVSManufacturing precision

Solution Approach 1:

The patent transforms 2D fisheye images into 3D spherical representations, then projects them onto a common plane. This dimensional transformation allows the system to preserve the wide field of view while correcting distortion through geometric projection mathematics, effectively resolving the contradiction between large FOV and image accuracy

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

Solution Approach 2:

The system changes the projection parameters by modeling each camera's view as a 3D shape and transforming coordinates from spherical to planar space. This parameter transformation enables distortion correction while maintaining the expansive field of view captured by wide-angle lenses

Inventive Principle:
Principle #35Parameter changes

2Area of stationary object

If images from multiple cameras with different tilt angles are combined, then the coverage area is improved, but alignment accuracy deteriorates

Engineering Contradiction:
Improvecoverage areaVSAvoidalignment accuracy
Core Design Contradiction:
Area of stationary objectVSMeasurement precision

Solution Approach 1:

By elevating 2D images to 3D spherical models, the system creates a common geometric framework that naturally accommodates different camera tilt angles. The 3D-to-2D projection process then automatically aligns all views to a common plane, resolving alignment issues while preserving comprehensive coverage

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

Solution Approach 2:

The patent introduces 3D spherical representations as an intermediary between the raw fisheye images and the final combined view. This intermediate 3D model serves as a common reference frame that facilitates accurate alignment of images from cameras with different orientations, thereby maintaining both coverage and precision

Inventive Principle:
Principle #24Intermediary (Mediator)

3Ease of manufacture

If traditional panorama stitching is used for wide-angle images, then the processing simplicity is improved, but the effectiveness in handling large FOV and distortion deteriorates

Engineering Contradiction:
Improveprocessing simplicityVSAvoideffectiveness in handling large FOV and distortion
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The patent replaces traditional 2D panorama stitching with a 3D spherical projection approach. This dimensional elevation provides a mathematically rigorous framework for handling wide-angle distortion, achieving both effectiveness in distortion correction and reasonable processing complexity through systematic coordinate transformations

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

Data Source

PatentEP3304498B1Systems and methods for producing a combined view from fisheye cameras
Publication Date: 2020.08.26 QUALCOMM INC
  • EP3304498B1 patent drawingFigure 1
  • EP3304498B1 patent drawingFigure 2
  • EP3304498B1 patent drawingFigure 3

AI summary

An electronic device is described. The electronic device includes a processor. The processor is configured to obtain images from a plurality of cameras. The processor is also configured to project each image to a respective 3-dimensional (3D) shape for each camera. The processor is further configured to generate a combined view from the images.