Accelerated PatchMatch Optical Flow Computation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Optical flow computations are resource-intensive for computing devices, consuming significant processing and memory resources, particularly when using the PatchMatch algorithm for applications like visual surveillance and augmented/virtual reality.
Innovation Solution
An accelerated and efficient PatchMatch algorithm is implemented, which performs raster scans in parallel, executes a propagation process based on optical flow estimates, and applies filtering and scaling to generate optical flow between successive image frames, reducing resource consumption by leveraging metrics like data, rigidity, and constraint metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the PatchMatch algorithm is applied to compute optical flow, then optical flow computation performance is improved, but processing and memory resources are consumed
Solution Approach 1:
The patent divides the image into patches and processes them independently through raster scans, enabling parallel computation. Each patch is handled as a separate unit, allowing the system to compute optical flow for multiple regions simultaneously, thereby improving productivity while managing resource consumption through selective processing.
Solution Approach 2:
The patent performs preliminary actions by pre-computing optical flow estimates for patches and using these estimates to guide subsequent propagation processes. This allows the system to make initial computations efficiently and use these results to accelerate further processing, reducing overall resource requirements while maintaining high computation performance.
2Speed
If raster scans are performed in parallel, then computation speed is improved, but complexity of the algorithm increases
Solution Approach 1:
The patent segments the raster scan process into multiple independent scan directions (horizontal and vertical) that can execute in parallel. Each scan direction processes specific patches independently, enabling concurrent computation without requiring complex coordination between all possible scan paths, thus achieving speed improvement with manageable complexity.
Solution Approach 2:
The patent introduces a new dimension to the computation by processing raster scans in multiple directions simultaneously (horizontal and vertical). This dimensional approach allows parallel execution of scan operations, improving computation speed while organizing complexity through structured multi-directional processing rather than requiring complex sequential algorithms.
Data Source
AI summary
A method for generating an optical flow for a plurality of successive image frames includes executing an initialization process by performing a plurality of raster scans of a patch of pixels in one or more of the plurality of successive image frames in parallel. The plurality of raster scans of the patch of pixels includes a plurality of optical flow estimates between the plurality of successive image frames. The method includes executing a propagation process based on the plurality of optical flow estimates between the plurality of successive image frames. Executing the propagation process includes propagating the plurality of optical flow estimates for one or more neighboring pixels associated with the patch of pixels. The method includes executing a search process by identifying one or more offsets based on the plurality of optical flow estimates for the one or more neighboring pixels associated with the patch of pixels.