Monocular Visual SLAM Pose Estimation via Homography Decomposition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Monocular visual SLAM systems require users to perform specific movements to capture images with sufficient parallax, making it difficult for untrained users to initialize the system effectively, as existing methods rely heavily on user operation and large-magnitude motion.

Innovation Solution

The method determines the pose of an image capturing device by estimating the position information of feature points in subsequent image frames using the position information from previous frames, allowing for quick and efficient initialization without requiring specific user movements, utilizing homography matrix decomposition and local bundle adjustment optimization to update depth information and camera pose.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional SfM technology is used for scene reconstruction, then accurate three-dimensional structure can be obtained, but the system requires large parallax changes and specific user movements that are difficult for untrained users to perform

Engineering Contradiction:
Improvescene reconstruction accuracyVSAvoiduser operation difficulty
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The patent changes the fundamental parameters of the initialization approach by abandoning the traditional SfM requirement for large parallax and instead using small-magnitude motion assumptions with homography matrix decomposition. This allows the system to work with minimal user movement while still achieving accurate scene reconstruction through iterative optimization of depth information and camera pose.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

Instead of requiring large parallax changes to initialize the system (traditional approach), the patent inverts the approach by assuming small parallax changes and using homography decomposition to initialize the scene. This reversal makes the system usable by untrained users who cannot perform specific movement patterns, while maintaining reconstruction accuracy through subsequent optimization steps.

Inventive Principle:
Principle #13The other way round (Inversion)

2Reliability

If large-magnitude motion is required for initialization, then sufficient parallax for SfM can be achieved, but the initialization time increases and user success rate decreases

Engineering Contradiction:
Improveinitialization success rateVSAvoidinitialization time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary action by using homography matrix decomposition to quickly initialize the scene structure and camera pose before refining with bundle adjustment. This preliminary initialization works with minimal user movement, ensuring high success rates even for untrained users, and reduces initialization time by avoiding the need for large-magnitude motion sequences.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system performs self-service initialization through automatic homography decomposition and iterative optimization without requiring user intervention to perform specific movements. The algorithm automatically adapts to the available small-magnitude motion in the video sequence, making the initialization process reliable and fast without user training.

Inventive Principle:
Principle #25Self-service

3Ease of operation

If homography matrix decomposition is used for initialization, then small-magnitude motion can be utilized, but depth information accuracy may be compromised without further optimization

Engineering Contradiction:
Improvemotion requirementVSAvoiddepth information accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent implements feedback through iterative bundle adjustment optimization that uses the homography-initialized scene as a starting point and refines depth information and camera pose based on actual feature point observations. This feedback loop continuously improves measurement precision while maintaining the ease of operation benefits from small-magnitude motion acceptance.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The system dynamically transitions from the static homography initialization to dynamic iterative optimization, adapting the scene structure and depth information as more video frames are processed. This dynamic refinement ensures that depth accuracy improves over time while the system remains accessible to users performing only small-magnitude motions.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3786890B1Method and apparatus for determining pose of image capture device, and storage medium therefor
Publication Date: 2023.08.30 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • EP3786890B1 patent drawingFigure 1~2
  • EP3786890B1 patent drawingFigure 3
  • EP3786890B1 patent drawingFigure 4

AI summary

A method for determining the pose of an image capture device, comprising: acquiring multiple frames of image captured by an image capture device; extracting multiple matching feature points in the multiple frames of image and first position information of each of the matching feature points in each of the multiple frames of image; estimating second position information of each of the matching feature points in a current frame of image in the multiple frames of image using the extracted first position information of each of the matching feature points in a previous frame of image of the multiple frames of image; and determining the pose of the image capture device based on the first position information and second position information of each of the matching feature points in the current frame of image.