Sub-Integer Position Estimation in Video Processing Hardware
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computer vision frameworks often detect feature or object positions at integer pixel locations, which lack sufficient accuracy for subsequent processing, necessitating the implementation of local maxima sub-integer position estimation in video processing.
Innovation Solution
The solution involves a hardware-based approach that estimates sub-integer positions of local maxima by applying a second-order polynomial interpolation using simple hardware circuitry, allowing for sub-pixel resolution and fractional bit results, and is implemented as part of a computer vision system that includes a processor circuit, coprocessor circuit, and dynamic random access memory.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If integer pixel location detection is used, then device complexity is reduced, but measurement precision deteriorates
Solution Approach 1:
The patent changes the parameter of position representation from integer pixel locations to sub-integer positions using fractional bits. By representing positions with fractional precision (e.g., 2 fractional bits for quarter-pixel accuracy) and using parameter transformations like gradient-based refinement, the system achieves higher measurement precision without proportionally increasing device complexity
Solution Approach 2:
The patent segments the position estimation process into discrete steps: initial integer pixel detection, gradient calculation, and iterative refinement using fractional bit representations. This segmentation allows the complex task of sub-integer estimation to be broken down into manageable hardware operations that can be implemented efficiently
2Measurement precision
If sub-integer position estimation is implemented, then measurement precision is improved, but computational complexity increases
Solution Approach 1:
The patent replaces complex continuous mathematical optimization with discrete hardware-based gradient calculations and lookup tables. By substituting mechanical/computational complexity with dedicated hardware circuits that perform gradient estimation and fractional position calculation in parallel, the system achieves sub-integer precision without proportional increases in computational burden
Solution Approach 2:
The patent implements partial refinement by performing gradient-based position estimation only in regions of interest where local maxima are detected, rather than processing the entire image. This partial action approach achieves sufficient precision for tracking applications without the excessive computational cost of full-image sub-integer analysis
Data Source
AI summary
An apparatus include a first circuit and a second circuit. The first circuit may be configured to (i) receive a plurality of sample values from a plurality of images in a video signal and (ii) estimate a plurality of positions of a plurality of maximum values in the images. Each estimation may operate on the sample values in a respective local region oriented parallel to an axis. The second circuit may be configured to track the positions of the maximum values in the images.