Head-worn Display Corner Detection via Two-Stage Resolution Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current head-worn image display devices face challenges in efficiently detecting corners in images for localization purposes, as existing corner detection methods require significant computational resources and may take too long to complete, affecting real-time processing and localization accuracy.
Innovation Solution
The proposed solution involves a processing unit in a head-worn image display device that obtains a first image with a high resolution and converts it into a second image with a lower resolution, using Harris corner detection to detect corners in the second image, followed by non-maxima suppression and spatial binning to refine corner positions, thereby improving localization accuracy with reduced computational demands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If corner detection is performed on high-resolution images, then localization accuracy is improved, but processing time increases and real-time performance deteriorates
Solution Approach 1:
The patent segments the corner detection process into two stages: first performing detection on downsampled low-resolution images to identify candidate corner regions, then refining detection only in those specific regions using high-resolution images. This segmentation allows the system to achieve high localization accuracy while maintaining real-time processing performance by limiting expensive high-resolution processing to only necessary image regions.
Solution Approach 2:
The patent applies preliminary downsampling and coarse corner detection on low-resolution images before performing detailed corner detection on high-resolution images. This preliminary action identifies promising regions of interest, allowing the system to prepare and focus computational resources in advance, thereby reducing overall processing time while maintaining accuracy.
2Measurement precision
If corner detection is performed on high-resolution images, then detection accuracy is improved, but computational resources increase
Solution Approach 1:
The patent divides the image processing task into coarse detection on downsampled images and refined detection on original-resolution images, applying computational intensity only where needed. This segmentation dramatically reduces overall computational resource consumption while preserving detection accuracy in critical regions.
Solution Approach 2:
The patent applies full high-resolution corner detection only to selected regions of interest identified from low-resolution preprocessing, rather than applying it uniformly across the entire image. This partial action approach maintains detection accuracy where necessary while reducing total computational resource usage.
3Productivity
If real-time corner detection is implemented, then processing speed is improved, but localization accuracy may deteriorate
Solution Approach 1:
The patent implements a two-stage detection pipeline where rapid low-resolution processing provides real-time responsiveness, followed by targeted high-resolution refinement that enhances accuracy without sacrificing overall processing speed. This segmentation enables the system to meet real-time requirements while maintaining high localization accuracy.
Solution Approach 2:
The patent maintains continuous processing by performing parallel downsampling and coarse detection while preparing for subsequent high-resolution refinement. This continuous action ensures real-time performance is maintained throughout the pipeline rather than introducing idle waiting periods.
Data Source
AI summary
An apparatus configured for head-worn by a user, includes: a screen configured to present graphics for the user; a camera system configured to view an environment in which the user is located; and a processing unit coupled to the camera system, the processing unit configured to: obtain a first image with a first resolution, the first image having a first corner, determine a second image with a second resolution, the second image having a second corner that corresponds with the first corner in the first image, wherein the second image is based on the first image, the second resolution being less than the first resolution, detect the second corner in the second image, determine a position of the second corner in the second image, and determine a position of the first corner in the first image based at least in part on the determined position of the second corner in the second image.