Binocular Image Stitching Using Rotation-Based Horizontal Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing binocular image stitching methods require significant hardware resources, are costly due to depth camera usage, suffer from poor real-time performance in cloud-based approaches, and involve high computational overhead in feature point matching, making them unsuitable for scenarios with strict real-time requirements or unreliable network conditions.
Innovation Solution
A method that optimizes binocular image stitching by utilizing hardware resources efficiently through horizontal corrections, 90-degree rotations, and image fusion, splitting corrections into two separate stages to fully utilize both hardware and software resources, reducing reliance on GPUs and minimizing ghosting and dislocation phenomena.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If binocular image stitching is performed using traditional methods with depth cameras or cloud-based processing, then comprehensive image information and broader field of view are achieved, but hardware costs increase, real-time performance deteriorates, and network dependency increases
Solution Approach 1:
The patent segments the image stitching process into distinct modular stages: horizontal correction module, rotation module, and fusion module. Each module processes specific aspects of the binocular images independently, enabling parallel processing and optimizing resource utilization. This segmentation allows real-time processing by breaking down the complex stitching operation into manageable, efficient steps that can be executed sequentially or in parallel depending on hardware capabilities.
Solution Approach 2:
The patent introduces a rotation dimension by rotating corrected images by 90 degrees before fusion. This dimensional transformation allows the system to utilize horizontal correction algorithms more effectively and enables the fusion module to process images in an optimized orientation, improving both processing efficiency and output quality while maintaining real-time performance requirements.
2Manufacturing precision
If feature point matching algorithms are used for binocular image stitching, then accurate alignment is achieved, but computational overhead increases significantly
Solution Approach 1:
The patent applies preliminary horizontal correction to both left and right binocular images before performing any matching or fusion operations. This pre-processing step corrects horizontal distortions and aligns the images in the horizontal direction, significantly reducing the computational complexity of subsequent matching operations. By performing this correction in advance, the system achieves accurate alignment with much lower computational overhead during the actual stitching process.
3Productivity
If GPU resources are heavily utilized for image stitching, then processing speed is improved, but hardware performance requirements and costs increase
Solution Approach 1:
The patent replaces complex GPU-based parallel processing with a streamlined CPU-based processing pipeline that utilizes efficient algorithms and data structures. The horizontal correction module and fusion module are designed to maximize CPU utilization through optimized memory access patterns and reduced data transfer requirements. This substitution maintains high processing speeds while significantly reducing hardware performance requirements, making the system accessible on standard computing platforms without specialized graphics hardware.
Data Source
AI summary
An image processing method, an image processing device, and an electronic device are provided. The image processing method includes: acquiring images to be processed based on captured images; correcting the images to be processed in a horizontal direction to obtain first corrected images; rotating the first corrected images by 90 degrees along a first direction to obtain second corrected images; and correcting the second corrected images in the horizontal direction to obtain third corrected images. In the present technical solution, the software algorithm is optimized for binocular image stitching while fully utilizing hardware resources, thereby effectively reducing hardware performance requirements.