Monocular Camera Pose Estimation via Neural Scale Adjustment

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Monocular visual SLAM algorithms face scale ambiguity issues in accurately estimating camera pose changes between images, leading to inaccurate movement prediction despite accurate feature point matching.

Innovation Solution

A neural network is employed to estimate camera position changes between images captured at different times, which are then used to adjust the scale of initial pose variation estimates, thereby minimizing errors and resolving scale ambiguity in SLAM operations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If monocular visual SLAM algorithms perform feature point matching between images, then the initial pose variation estimation is obtained, but scale ambiguity arises leading to inaccurate camera movement prediction

Engineering Contradiction:
Improvecamera pose estimation accuracyVSAvoidscale ambiguity
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent introduces an intermediary component (scale adjustment module) that mediates between the feature point matching results and the final pose estimation. This module uses the estimated position change from neural network as a reference to adjust the scale of the initial pose variation, resolving the scale ambiguity without requiring additional sensors or complex modifications to the existing SLAM pipeline.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent changes the scale parameter of the initial pose variation estimation by using the estimated position change from the neural network as a reference. This parameter adjustment resolves the scale ambiguity issue while maintaining the computational efficiency of monocular visual SLAM algorithms.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If neural network is used to estimate camera position change, then scale ambiguity is reduced, but computational complexity increases

Engineering Contradiction:
Improveposition change estimation accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies partial action by using the neural network only for estimating position change magnitude, while the existing feature point matching algorithm continues to handle pose variation estimation. This partial application of neural network reduces the overall computational burden compared to replacing the entire SLAM pipeline.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The patent segments the pose estimation task into two parts: (1) initial pose variation estimation through feature point matching, and (2) scale adjustment using neural network-based position change estimation. This segmentation allows each component to be optimized independently, balancing accuracy and computational complexity.

Inventive Principle:
Principle #1Segmentation

3Measurement precision

If scale adjustment is performed based on neural network estimation, then camera movement prediction accuracy improves, but processing time increases

Engineering Contradiction:
Improvecamera movement prediction accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary action by estimating the position change using the neural network before the scale adjustment step. This allows the scale adjustment to be performed efficiently using pre-computed position change information, reducing the overall processing time compared to iterative optimization approaches.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4383198A1Device and method with camera pose estimation
Publication Date: 2024.06.12 SAMSUNG ELECTRONICS CO LTD
  • EP4383198A1 patent drawingFigure 1
  • EP4383198A1 patent drawingFigure 2
  • EP4383198A1 patent drawingFigure 3

AI summary

Disclosed are a device and method with pose estimation. A method of operating an electronic device includes: obtaining monocular images captured at different respective points in time by a same camera; estimating, based on a neural network, a position change of the camera corresponding to the time between the points in time at which the images were captured; and estimating a position and/or orientation of the camera based on the estimated position change of the camera and based on the images.