3D Model Reconstruction via Image Similarity Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional three-dimensional model reconstruction and fusion methods in automatic-driving applications, such as autonomous parking or cruising, face significant computational burdens and resource consumption due to large data volumes, leading to inefficiencies and user experience issues.
Innovation Solution
A parallel three-dimensional reconstruction and fusion method is introduced, where image sequences are grouped based on similarity measurements, allowing for the creation of an original three-dimensional model using a basic image sequence and subsequent refinement using other image sequences, optimizing data processing and reducing computational complexity from exponential to linear growth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional three-dimensional model reconstruction methods are used with large volumes of driving record images, then the completeness and accuracy of the three-dimensional model is improved, but the computational burden and system resource consumption increase exponentially
Solution Approach 1:
The patent segments the large set of driving record images into multiple subsets based on similarity measurements. Each subset is processed independently to create intermediate three-dimensional models, which are then fused into a final comprehensive model. This segmentation reduces the computational complexity from exponential to linear growth while maintaining model accuracy through systematic fusion of partial results.
2Reliability
If all driving record images are processed for three-dimensional model reconstruction, then the completeness of the model is improved, but the processing time increases significantly
Solution Approach 1:
The patent performs preliminary actions by calculating similarity measurements between images before the actual three-dimensional reconstruction process. Based on these pre-computed similarities, images are grouped into subsets that can be processed in parallel. This preliminary organization enables efficient resource allocation and significantly reduces the overall reconstruction time while ensuring all necessary images are eventually processed for complete model coverage.
3Measurement precision
If the amount of driving record image data is increased, then the quality of the three-dimensional model is improved, but the system resource occupation increases exponentially
Solution Approach 1:
The patent divides the large dataset into multiple smaller subsets based on image similarity, allowing each subset to be processed with moderate resource requirements. The segmentation strategy ensures that no single processing task requires exponential resources, while the collective processing of all subsets delivers high-quality comprehensive model results through systematic fusion of intermediate models.
Solution Approach 2:
The patent processes images in partial batches organized by similarity groups rather than attempting to process all images simultaneously. This partial action approach allows the system to manage resource occupation at linear levels while still achieving complete model coverage through multiple processing passes and fusion operations.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
According to an exemplary embodiment of the present disclosure, a method and apparatus for processing an image are provided, which may be used in the field of automatic-driving or autonomous parking. The method may include: acquiring a set of image sequences, the set of image sequences including a plurality of image sequence subsets divided according to similarity measurements between image sequences, each image sequence subset including a basic image sequence and other image sequence, wherein a first similarity measurement corresponding to the basic image sequence is greater than or equal to a first similarity measurement corresponding to the other image sequence; creating an original three-dimensional model using the basic image sequence; and creating a final three-dimensional model using the other image sequence based on the original three-dimensional model. With the above method, the efficiency of creating the three-dimensional model may be improved and the overhead of creating the three-dimensional model may be reduced.