Scenery Model Construction via Sub-model Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing structure from motion (SfM) methods face challenges in maintaining accuracy and efficiency when constructing and updating three-dimensional models of scenery from multiple images, due to issues like drifting and error accumulation, especially when incorporating additional camera information.
Innovation Solution
A computer-implemented method that divides the model into sub-models, each created from overlapping images, allowing for separate updating and merging of camera positions and angles, enabling the use of additional information like GPS and compass data to improve accuracy without requiring complete reconstruction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If incremental SfM is used to add images one by one to reconstruct scenery, then the model can be built step by step, but error accumulation and drifting occur reducing accuracy
Solution Approach 1:
The patent divides the complete SfM model into multiple sub-models, each constructed from a specific subset of images. This segmentation allows independent construction and updating of sub-models, preventing error accumulation across the entire model while maintaining overall accuracy. Each sub-model can be updated separately with new images without affecting the entire reconstruction.
2Measurement precision
If non-incremental batch SfM is used to handle all images simultaneously, then accuracy can be improved, but computational complexity increases significantly
Solution Approach 1:
The patent segments the batch SfM process into multiple sub-models, each processed independently. This allows the system to maintain batch processing accuracy for each sub-model while reducing overall computational complexity by avoiding the need to reprocess the entire dataset simultaneously. Only relevant sub-models are updated when new images arrive.
Solution Approach 2:
The patent introduces dynamic updating of sub-models based on incoming images. The system can switch between incremental updates for individual sub-models and batch processing for complete model reconstruction, adapting the computational approach based on the current state and available resources.
3Measurement precision
If additional camera information like GPS and compass data is incorporated, then positional accuracy improves, but the system becomes more complex and sensitive to outliers
Solution Approach 1:
The patent segments camera information processing into sub-models, where each sub-model incorporates GPS and compass data independently. This segmentation isolates outlier effects to specific sub-models rather than propagating through the entire system, maintaining overall accuracy while managing complexity through modular processing.
Data Source
Figure 1~3
Figure 4~5
Figure 6
AI summary
A method in constructing a model of a scenery comprises: receiving (200) a plurality of images depicting the scenery, wherein each image is acquired by a camera (104) from a position at a rotational angle; creating (202) a plurality of sub-models, wherein each sub-model is created by: selecting (400) a basis image; determining (402) images related to the basis image; and determining (404) the sub-model based on the basis image and the related images, wherein the sub-model defines relative positions and rotational angles of the camera (104) for the basis image and the related images and the sub-model further defines three-dimensional structures of objects depicted in the plurality of images; and separately merging (204) the defined relative positions and rotational angles of cameras (104) for the plurality of sub-models.