Homography-Based Image Compression for Vehicle Cameras
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image compression techniques, such as H.264, struggle with accurately calculating motion vectors for images captured by vehicle-mounted cameras, especially when objects at varying distances are involved, leading to reduced compression efficiency due to uniform enlargement or reduction of reference images, which does not account for distance-related variations.
Innovation Solution
An image compression method that sets segmented regions in the reference image, with the central region corresponding to the camera's optical axis, and enlarges peripheral regions more than the central region, generating a predictive image for improved concordance with the image to be encoded, thereby enhancing compression efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If uniform enlargement or reduction of reference image is applied, then processing simplicity is maintained, but motion vector calculation accuracy deteriorates for objects at varying distances
Solution Approach 1:
The reference image is divided into multiple regions (e.g., near-field region and far-field region) based on distance from the optical axis. Each region is then enlarged or reduced by a different scale factor, allowing differential processing that accounts for varying object distances while maintaining manageable processing complexity through systematic regional division.
Solution Approach 2:
Different regions of the reference image are assigned different enlargement/reduction ratios according to their distance characteristics. The near-field region uses a larger scale factor while the far-field region uses a smaller scale factor, creating local quality variations that match the physical characteristics of objects at different distances and improve motion vector accuracy.
2Productivity
If differential value is calculated for images with zooming operations, then compression encoding is performed, but large differential values are generated reducing compression efficiency
Solution Approach 1:
The reference image is pre-enlarged or reduced according to the zooming operation characteristics before motion compensation. By applying the appropriate scale factor to the reference image in advance, the differential calculation between reference and current frames produces smaller values, improving compression efficiency while maintaining encoding functionality.
3Measurement precision
If global motion compensation is applied, then accuracy for handling camera movements is improved, but compatibility with H.264 standard is reduced
Solution Approach 1:
The global motion compensation technique is segmented into regional operations within the H.264 framework. By dividing the image into regions and applying different transformation parameters to each region, the patent adapts global motion compensation concepts to work within the block-based structure of H.264, maintaining both accuracy and compatibility.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image compression method, performed by a computer, includes: acquiring image data of an image captured by an imaging device mounted to a vehicle; setting a segmented region in a reference image corresponding to an image to be compressed and encoded; enlarging the reference image so that a magnification rate of a region including the periphery of the reference image is greater than a magnification rate of a region including a center point of the reference image in the segmented region, the center point being a contact point of an optical axis of the image incident to the imaging device and an imaging element in the imaging device; generating a predictive image in accordance with the enlarged reference image; and compressing and encoding image data of the image to be encoded in accordance with the generated predictive image.