3D Point Cloud Densification With Reliability-Guided Image Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 3D point cloud generation techniques face challenges in achieving computational efficiency and accuracy, with exhaustive matching algorithms being too demanding and sequential matching algorithms often failing to produce reliable sparse point clouds for dense reconstruction.
Innovation Solution
A method and device that uses sequential matching to establish correspondence between images, estimates camera poses, determines reliability values, and generates dense point clouds by densifying sparse clouds only when reliability meets a threshold, switching to exhaustive matching if necessary, and utilizing help data or re-scanning for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If exhaustive matching algorithm is used to generate sparse 3D point cloud, then measurement precision and reliability are improved, but computational time and processing complexity increase significantly
Solution Approach 1:
The system dynamically switches between sequential matching mode (for time efficiency) and exhaustive matching mode (for high reliability) based on real-time reliability assessment of the sparse point cloud. This dynamic adaptation allows the system to optimize the balance between computational time and reliability depending on the specific reconstruction task requirements.
Solution Approach 2:
The system changes the matching algorithm parameter (from sequential to exhaustive) based on the reliability threshold. When reliability falls below the threshold, the system switches to exhaustive matching; otherwise, it continues with sequential matching. This parameter change enables flexible adaptation to different reliability requirements.
2Productivity
If sequential matching algorithm is used to generate sparse 3D point cloud, then computational efficiency is improved, but measurement precision and reliability deteriorate
Solution Approach 1:
The system implements a feedback mechanism that continuously assesses the reliability of the generated sparse point cloud and uses this information to determine whether to switch to exhaustive matching. This feedback loop ensures that computational efficiency is maintained while reliability requirements are met through adaptive algorithm selection.
3Manufacturing precision
If dense 3D point cloud is generated from unreliable sparse point cloud, then visualization and processing quality improve, but computational resources are wasted on unsuccessful reconstructions
Solution Approach 1:
The system performs a preliminary reliability assessment of the sparse point cloud before proceeding to dense reconstruction. This preliminary action filters out unreliable reconstructions early in the process, preventing wasteful computational resources from being spent on generating dense point clouds from poor sparse foundations.
Data Source
Figure 1~3
Figure 4
Figure 5
AI summary
There is provided techniques for generating a dense 3D point cloud of a scene. A method is performed by an image processing device (400). The method comprises performing (S102) sequential matching for a set of images, obtained from an image scan of the scene, where each image has a color value and a depth value, to establish a correspondence between consecutive images in the set of images, to generate a sparse 3D point cloud from the images, and to estimate one camera pose value for each image. The method comprises determining (S104) one reliability value for each camera pose value. The method comprises generating (S114a) the dense 3D point cloud, when a lowest value of all determined reliability values is larger than a threshold reliability value, by densification of the sparse 3D point cloud and using the estimated camera pose values.