3D Reconstruction for Vehicle Using Sparse SfM

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional 3D reconstruction techniques for vehicle environments are computationally expensive and not suitable for real-time applications, especially with wide-angle cameras, due to reliance on multiple cameras, complex calibration processes, and uncertainties in vehicle odometry data, making them impractical for automated driving assistance systems.

Innovation Solution

A method and system using a single camera with vehicle odometry data to perform sparse Structure from Motion (SfM) by selecting optimal subsets of tracked image points based on geometric quality metrics, such as reprojection error, to reconstruct 3D coordinates efficiently, reducing computational load and improving accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If stereo vision-based systems with multiple cameras are used for 3D reconstruction, then measurement precision is improved, but device complexity and computational cost increase

Engineering Contradiction:
Improve3D reconstruction accuracyVSAvoidnumber of cameras and calibration requirements
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the 3D reconstruction process into two distinct phases: an offline calibration phase that establishes camera parameters and a online reconstruction phase that uses pre-computed matrices. This segmentation allows the complex calibration to be performed once rather than continuously, reducing real-time computational burden while maintaining accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs camera calibration and computes essential matrices in advance during an offline phase. These pre-computed parameters are then reused during real-time 3D reconstruction, eliminating the need for repeated calibration calculations and significantly reducing online computational complexity.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If Structure from Motion with dense feature points is used for 3D reconstruction, then measurement precision is improved, but productivity decreases due to computational expense

Engineering Contradiction:
Improve3D reconstruction accuracyVSAvoidreal-time processing capability
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent extracts and utilizes only the essential geometric information (essential matrices) from the calibration process, discarding the computationally intensive dense feature matching requirements. By working with extracted key parameters rather than complete dense reconstructions, the system achieves real-time performance.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the problem parameters from dense 3D point cloud reconstruction to sparse point reconstruction using pre-computed essential matrices. This parameter transformation reduces computational complexity from O(N^2) to O(N) where N is the number of feature points, enabling real-time processing.

Inventive Principle:
Principle #35Parameter changes

3Ease of operation

If vehicle odometry data is used for camera location estimation, then ease of operation is improved, but measurement precision deteriorates due to uncertainties and noise

Engineering Contradiction:
Improveavailability of motion dataVSAvoidcamera location accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent introduces visual features from image sequences as an intermediary to bridge the gap between odometry data and accurate camera positioning. By tracking visual features across frames and combining them with odometry, the system compensates for odometry errors and achieves more precise location estimation.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent implements a feedback mechanism where tracked visual features are used to correct and refine camera position estimates derived from odometry. The visual information provides continuous feedback that compensates for cumulative odometry errors, improving overall positioning accuracy.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP3293700B13D reconstruction for vehicle
Publication Date: 2019.11.13 PANASONIC AUTOMOTIVE & IND SYST EURO GMBH
  • EP3293700B1 patent drawingFigure 1
  • EP3293700B1 patent drawingFigure 2
  • EP3293700B1 patent drawingFigure 3(a)~3(f)

AI summary

The present invention relates to a method and system for reconstructing a portion of the environment of a moving vehicle based on an image sequence captured during vehicle movement at a predetermined frame rate and vehicle odometry data obtained from conventional automotive equipment, via a vehicle bus. The reconstruction is performed on the basis of a sparse Structure from Motion approach, wherein an optimal subset of tracked image points for triangulation is determined on the basis of optimizing a geometric quality metric.