Panoramic Stitching Parallax Minimization via Depth Map Projection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Stitching artifacts due to parallax errors occur when near-field images from cameras with different centers of projection are merged to form panoramic views, especially in environments with short focal lengths and complex scenes like office buildings, leading to inaccuracies in representing the entire area.
Innovation Solution
A method that estimates the depth of each pixel corresponding to a scene point using a ranging device and generates a spherical coordinate system to partition the scene into volumetric pixels, allowing for accurate projection and mapping of pixels from individual cameras onto a panoramic image, thereby minimizing parallax errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If multiple cameras with different centers of projection are used to capture near-field images for panoramic stitching, then the field of view coverage is improved, but stitching artifacts due to parallax errors increase
Solution Approach 1:
The patent performs preliminary depth estimation and parallax compensation calculations before the actual stitching process. By pre-computing the depth map from multiple views and calculating the necessary pixel shifts to compensate for parallax, the system prepares correction data in advance that is then applied during stitching to eliminate artifacts and improve accuracy
Solution Approach 2:
The patent introduces an intermediate depth map and parallax compensation layer between the raw camera images and the final stitched panorama. This intermediary representation captures the three-dimensional depth information and uses it to calculate correction transformations that account for the different centers of projection, thereby mediating the stitching process to reduce artifacts
2Measurement precision
If the focal length is decreased for near-field imaging, then the ability to capture detailed close-up features is improved, but parallax errors become more significant
Solution Approach 1:
The patent changes the parameter representation from two-dimensional image coordinates to three-dimensional depth-aware coordinates. By estimating depth for each pixel and using this depth information to guide the stitching process, the system adapts the stitching parameters based on the actual distance to scene points, thereby compensating for the increased parallax effects that result from using shorter focal lengths
3Adaptability or versatility
If cameras are positioned at different locations on a common platform to achieve 360 degree coverage, then the panoramic view completeness is improved, but the number of centers of projection increases causing more stitching artifacts
Solution Approach 1:
The patent creates a universal depth map representation that serves all cameras simultaneously. By estimating depth from multiple views and creating a unified three-dimensional model of the scene, this single depth representation is then used to guide the stitching process for all cameras with different centers of projection, thereby providing a universal solution that handles multiple projection centers without increasing complexity
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enables the creation of high-resolution panoramic images with minimized parallax, ensuring accurate representation of the scene by spatially shifting pixels based on depth data, resulting in improved image quality and reduced stitching artifacts.
Implementation Method 1
estimating a depth (distance) of each pixel corresponding to a scene point on a scene surface... A plurality of scan lines from a ranging device positioned on a camera assembly may be identified
Data Source
AI summary
Systems, methods and articles of manufacture for minimizing parallax based on depth map generation are described herein. Embodiments of the present disclosure relate to cameras positioned on a camera assembly that capture digital photographic images. The digital photographic images are stitched together into a panoramic image. In order to minimize the parallax while stitching, the procedure includes estimating a distance of a pixel corresponding to a scene point on the scene surface from a panoramic center of projection positioned on the camera assembly to the scene surface. A depth map is generated that provides the estimated depth of the pixel. Based on the estimated depth, the pixel may be projected onto each camera. A panoramic pixel corresponding to the scene point may be obtained from the values of each projected pixel. The panoramic pixel may then be mapped onto the panoramic image with parallax minimized.