Camera Pose Tracking via Homography Complementary Filtering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current augmented reality technologies face challenges in determining camera pose information with low precision and efficiency, especially when using natural images as markers, due to the lack of consideration for feature point changes and excessive match overheads during image layer processing.

Innovation Solution

A method involving feature point detection and homography calculation using a template image and a second image, with complementary filtering processing to enhance precision and stability, including downsampling the template image to generate a pyramid image, extracting feature points from each layer, and performing optical flow tracking to obtain camera pose information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If feature points of one image layer in the Marker image are matched with feature points in the current image, then the matching process is simple and fast, but the obtained camera pose information has relatively low precision

Engineering Contradiction:
Improvematching speedVSAvoidcamera pose precision
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The template image is divided into multiple image layers through downsampling to create a pyramid structure. Feature points are detected independently on each layer, allowing the system to capture features at different scales and improve matching precision without sacrificing too much speed.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a scale dimension by creating multiple image layers at different resolutions. This allows feature matching to occur across multiple scales, transforming a 2D feature matching problem into a multi-scale 3D feature space, thereby improving precision while maintaining computational efficiency through selective layer processing.

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

2Measurement precision

If a plurality of image layers of the Marker image are obtained and feature points of each image layer are matched with the feature points in the current image, then the camera pose precision is improved, but excessive match overheads are needed which is adverse to the running efficiency

Engineering Contradiction:
Improvecamera pose precisionVSAvoidrunning efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

Feature points are pre-detected on each image layer of the pyramid structure before the actual matching process. This preliminary detection organizes features hierarchically, allowing the matching algorithm to work with pre-processed feature data rather than raw pixels, significantly reducing computational overhead during runtime.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

Different image layers are processed with different levels of detail. Coarser layers provide overall structure and large-scale features, while finer layers provide detailed local features. This local quality differentiation allows efficient matching by using coarse layers for initial alignment and fine layers for precise refinement, reducing total matching overhead.

Inventive Principle:
Principle #3Local quality

3Adaptability or versatility

If natural images are used as template images, then the system is more versatile and easier to operate, but the feature point detection and matching becomes more complex and less reliable

Engineering Contradiction:
Improvetemplate image versatilityVSAvoidfeature detection complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The natural image template is segmented into multiple pyramid layers, each capturing different scale information. This segmentation transforms the complex continuous image into discrete hierarchical levels, making feature detection more systematic and reliable while maintaining versatility with natural images.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent employs dynamic feature detection that adapts to the content of natural images by processing multiple pyramid layers. The system dynamically selects and weights features from different layers based on their relevance, making the detection process more robust to variations in natural image content while maintaining versatility.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3579192B1Method, apparatus and device for determining camera posture information, and storage medium
Publication Date: 2023.10.25 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • EP3579192B1 patent drawingFigure 1~2
  • EP3579192B1 patent drawingFigure 3~4
  • EP3579192B1 patent drawingFigure 5~6

AI summary

Embodiments of this application disclose a method, an apparatus, and a device for determining camera pose information, and a storage medium. The method includes: obtaining a first image, a second image, and a template image; performing feature point detection on a first feature point of the template image and a second feature point of the second image, to obtain a first homography; determining a first target homography according to a first optical flow feature point in the first image and a second optical flow feature point in the second image, and determining a second homography according to the first target homography and a second target homography; and performing complementary filtering processing on the first homography and the second homography, to obtain camera pose information of the camera. In the embodiments of this application, complementary filtering processing may be performed on two homographies obtained in a camera pose tracking process, to obtain more precise camera pose information.