Image Recognition Device Region Candidate Calculation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image recognition techniques using machine learning face challenges in efficiently detecting joint positions in images due to high computational requirements and difficulties in defining clear bounding boxes for joints, which are not easily recognizable.
Innovation Solution
An image recognition device that employs a hardware processor to calculate and determine region candidates for object parts in images, using machine learning to estimate the size and position of these regions based on predetermined criteria, reducing computational load and enabling accurate detection even when the recognition target's range is unclear.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If likelihood maps are calculated pixel by pixel to detect joint positions, then measurement precision is improved, but productivity deteriorates due to large computational amount
Solution Approach 1:
The image is divided into multiple grids, and each grid is processed independently to generate region candidates. This segmentation reduces the computational burden compared to processing the entire image pixel by pixel, while still maintaining accurate joint position detection through subsequent processing steps.
Solution Approach 2:
The patent extracts only the necessary region candidates from the image grids that are likely to contain joints, rather than processing all pixels. This extraction approach maintains measurement precision by focusing computational resources on relevant areas while improving productivity by avoiding unnecessary calculations.
2Productivity
If bounding boxes are used for object recognition, then productivity is improved due to simplified processing, but measurement precision deteriorates for targets with unclear ranges such as joints
Solution Approach 1:
Instead of applying a uniform bounding box approach to all targets, the patent adapts the region representation to the local characteristics of each target. For joints with unclear ranges, it generates multiple region candidates with different sizes and shapes, allowing precise representation of each joint's specific characteristics while maintaining overall processing efficiency.
Solution Approach 2:
The patent dynamically adjusts the region candidates based on the detected object characteristics. For joints where the range is not clear, the system generates multiple candidate regions with varying parameters and selects the most appropriate one, rather than using fixed bounding boxes. This dynamic adaptation maintains both productivity and measurement precision.
3Measurement precision
If multiple region candidates are generated for each grid, then measurement precision is improved for ambiguous targets, but device complexity increases
Solution Approach 1:
The patent generates multiple region candidates for each grid, which may seem excessive, but this partial redundancy actually simplifies the overall system by avoiding the need for complex adaptive algorithms. The multiple candidates provide sufficient information for accurate joint detection without requiring sophisticated real-time adjustment mechanisms, thus improving measurement precision while controlling device complexity.
Data Source
AI summary
An image recognition device includes: a hardware processor that: conducts machine learning, to perform a first process of calculating a plurality of region candidates for a region showing part of an object captured in an image, and a second process of determining a size of each of the region candidates in accordance with the object captured in the image; and determines the region from among the region candidates, using a predetermined criterion.