Adaptive Subsampled Look-Up Table for Camera Geometric Distortion Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current geometric distortion correction techniques in camera systems, such as those used in Advanced Driver Assistance Systems, face inefficiencies due to high bandwidth requirements for memory access and descriptor data reading, especially when dealing with non-linear distortions and varying distortion levels across images.
Innovation Solution
The implementation of an adaptive subsampling method for look-up tables, where the full look-up table is divided into patches and further subdivided based on maximum difference thresholds, reducing the amount of data needed for correction while maintaining image quality, by iteratively downsampling correction data to create an adaptively subsampled look-up table.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If direct remapping approach is used for geometric distortion correction, then correction quality is maintained, but bandwidth requirement vastly inflates
Solution Approach 1:
The patent divides the image into multiple tiles and processes each tile independently. This segmentation allows the system to pre-fetch source data for entire tiles at once, reducing the total number of memory accesses compared to processing individual pixels, thereby reducing bandwidth requirements while maintaining correction quality through systematic processing of each segment
Solution Approach 2:
The patent implements pre-fetching of source data from external memory before the actual distortion correction processing. By preparing and storing the necessary source pixel data in advance in on-chip memory, the system reduces the bandwidth burden during the correction phase, as data is retrieved in bulk operations rather than individual pixel accesses
2Quantity of substance
If tile-based geometric correction is implemented, then bandwidth utilization is reduced, but processing overhead at tile boundaries increases
Solution Approach 1:
The patent merges adjacent tiles at their boundaries by allowing overlap in the pre-fetched data regions. This merging approach ensures that pixels at tile boundaries are processed consistently without requiring re-fetching, thereby reducing processing overhead while maintaining the bandwidth benefits of tile-based processing
Solution Approach 2:
The patent applies different processing strategies to different regions of the image. Tile interior pixels use efficient pre-fetched processing, while boundary pixels are handled with special consideration through overlapping regions. This local differentiation optimizes the balance between bandwidth utilization and processing overhead by applying appropriate methods to each region
3Measurement precision
If full look-up table is used for distortion correction, then correction accuracy is maintained, but descriptor data reading bandwidth increases significantly
Solution Approach 1:
The patent extracts only the essential correction parameters from the full look-up table that are needed for each tile's processing. By selecting and storing only the relevant descriptor data in on-chip memory rather than accessing the complete look-up table, the system maintains correction accuracy for the processed region while significantly reducing the bandwidth required for descriptor data reading
Data Source
AI summary
A camera system and method are disclosed. The camera includes a memory, a camera, and a processor. The memory stores an adaptively subsampled look-up table. The adaptively subsampled look-up includes varying levels of subsampling across the adaptively subsampled look-up table. The camera captures an image, and the captured image is distorted based on varying distortions within an optical system of the camera. The processor receives the captured image from the camera, corrects the distorted captured image based on the adaptively subsampled look-up table to create a correct image, provides the corrected image, and executes a safety feature based on the corrected image.