Image Registration via Collineation Warping Decomposition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for registering part of a current camera image with a template image are computationally expensive and memory-intensive, especially on mobile devices, due to the need for eight-dimensional non-linear optimization and excessive pre-computed transformations, limiting their feasibility for arbitrary camera positions and orientations.
Innovation Solution
The method decomposes the collineation warping function into a collineation warping function computed from device orientation measurements and an affine transformation, reducing computational complexity and memory requirements by using inertial sensors, accelerometers, and other sensors to determine device orientation and distance, allowing for efficient registration of images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If standard collineation warping function is used for image registration, then registration accuracy for arbitrary camera positions and orientations is maintained, but computational cost and memory usage increase significantly
Solution Approach 1:
The collineation warping function is decomposed into two separate functions: a perspective warping function that handles arbitrary camera positions and orientations, and an affine transformation function that handles remaining registration adjustments. This segmentation allows each function to be optimized independently, reducing overall computational complexity while maintaining registration accuracy.
Solution Approach 2:
The perspective warping function is computed in advance based on device orientation measurements from inertial sensors, creating a pre-computed transformation that accounts for arbitrary camera positions and orientations. This preliminary action reduces the complexity of the subsequent registration step, as the affine transformation only needs to handle fewer degrees of freedom.
2Measurement precision
If eight-dimensional non-linear optimization is performed for image registration, then accurate registration is achieved, but processing time and computational resources increase
Solution Approach 1:
The registration process is divided into two stages: first applying a pre-computed perspective warping function based on device orientation, then performing a simplified affine transformation registration. This segmentation reduces the dimensionality of the non-linear optimization from eight dimensions to fewer dimensions, significantly reducing processing time while maintaining accuracy.
Solution Approach 2:
The problem is transformed by changing parameters from the full collineation matrix parameters to a combination of device orientation measurements (from inertial sensors) and simplified affine transformation parameters. This parameter change reduces the complexity of the optimization problem and decreases processing time.
3Measurement precision
If pre-computed transformations are stored for all possible camera positions and orientations, then registration accuracy is maintained, but memory requirements become excessive
Solution Approach 1:
Instead of storing pre-computed transformations for all possible camera positions and orientations, the system computes a perspective warping function on-demand based on real-time device orientation measurements from inertial sensors. This preliminary computation based on actual sensor data eliminates the need for storing large numbers of pre-computed transformations, significantly reducing memory usage while maintaining accuracy.
4Adaptability or versatility
If full collineation warping is applied, then accurate registration for arbitrary camera positions and orientations is achieved, but computational complexity increases
Solution Approach 1:
The collineation warping function is segmented into a perspective warping component that handles arbitrary camera positions and orientations using device orientation measurements, and an affine transformation component that handles the remaining registration. This segmentation maintains adaptability to arbitrary camera positions while reducing the computational complexity of the affine transformation step.
Solution Approach 2:
Device orientation measurements from inertial sensors serve as an intermediary that bridges the gap between arbitrary camera positions and the registration process. By using these measurements to compute the perspective warping function, the system maintains versatility for arbitrary positions while reducing the complexity of the subsequent registration step.
Data Source
AI summary
The invention is related to a method for registering at least one part of a first image and of a second image, including the steps of providing a first image of the real environment and a coordinate system associated to the first image, a second image of the real environment, and measurements of orientation and/or distance of a capturing device with respect to the real environment while capturing the second image. A collineation warping function is then determined from the measurements associated to at least one part of the second image. The method further includes the steps of transforming the at least one part of the second image by applying the corresponding determined collineation warping function, determining a second warping function for registering the transformed at least one part of the second image and the at least one part of the first image by means of an image registration method, and registering the at least one part of the second image and first image using the collineation warping function and the second warping function.