Image Frame Alignment Using Downsampled Motion Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in efficiently processing high-resolution image frames due to significant computational requirements and time consumption for motion estimation, which affects the accuracy and speed of object detection and navigation.
Innovation Solution
The computing system downsizes or down-samples image frames to lower resolutions, determining alignment information between smaller sized frames to align higher resolution frames, thereby reducing computation complexity and improving processing speed and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion estimation techniques are performed using original high-resolution image frames, then detection accuracy is improved, but processing time and computational resources increase significantly
Solution Approach 1:
The patent divides the image processing task into two segments: first processing downsampled low-resolution versions of image frames to obtain initial alignment information, then applying this alignment information to the original high-resolution frames. This segmentation allows the computationally intensive motion estimation to be performed on smaller data while preserving the ability to achieve accurate alignment on the full-resolution images.
Solution Approach 2:
The patent performs preliminary processing on downsampled image frames before processing the original high-resolution frames. By pre-computing alignment information from the downsampled versions, the system prepares initial alignment vectors that guide subsequent processing of the full-resolution images, reducing the overall computational burden and processing time.
2Measurement precision
If motion estimation techniques are performed using original high-resolution image frames, then detection accuracy is improved, but computational resources required increase significantly
Solution Approach 1:
The patent segments the computational workload by separating processing of downsampled frames from original frames. The complex motion estimation algorithm is applied first to the smaller downsampled versions, generating alignment information that is then used to process the original high-resolution frames more efficiently, thereby reducing peak computational resource requirements.
Solution Approach 2:
The downsampled image frames serve as an intermediary representation that facilitates the alignment process. By computing alignment vectors from these simplified versions first, the system creates a computational bridge that makes subsequent processing of the full-resolution images more manageable and less resource-intensive.
3Productivity
If downscaling image frames is performed, then computation complexity is reduced, but image quality may deteriorate
Solution Approach 1:
The patent applies different quality requirements to different stages of processing: downsampled frames are processed with reduced quality requirements to enable fast computation, while the final alignment is applied to the full-resolution frames to ensure high image quality in the output. Each stage uses the appropriate level of detail for its specific purpose.
Data Source
AI summary
Described examples relate to an apparatus comprising a memory for storing image frames and at least one processor. The at least one processor may be configured to receive a plurality of image frames from an image capture device and downsize each of the plurality image frames to generate a plurality of versions of each image frame at a plurality of different sizes. The at least one processor may also be configured to determine alignment information for a first version of a first image frame. The alignment information may include a first alignment vector for identifying image data in a first version of a second image frame that corresponds to image data in the first version of the first image frame. Further, the at least one processor may be configured to determine a first initial alignment vector for identifying image data in a first version of a third image frame based on at least the first alignment vector.