Machine Vision Efficiency Estimation for Grouped Objects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image recognition technologies lack effective mechanisms for evaluating the efficiency of machine vision, particularly when multiple objects are grouped together in an image, as they fail to accurately score and distinguish the number of identified objects.
Innovation Solution
A method and system that involve obtaining an image with multiple objects, performing image recognition to generate prediction blocks, merging standard blocks corresponding to each object, and using the merged blocks and prediction blocks to generate evaluation information that reflects the prediction efficiency of the machine vision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If machine vision selects objects by a large range to handle grouped objects, then the coverage of detection is improved, but the precision of identifying individual objects deteriorates
Solution Approach 1:
The patent segments the detection process into two levels: first, a large-range detection identifies grouped objects as a whole; second, the grouped objects are segmented into individual objects for precise identification and counting. This segmentation resolves the contradiction by maintaining both broad coverage and fine-grained precision through hierarchical processing.
Solution Approach 2:
The patent introduces a new evaluation dimension by creating merged standard blocks that combine individual object standard blocks. This dimensional transformation allows the system to evaluate both group-level detection and individual object identification simultaneously, resolving the precision-coverage tradeoff by operating in multiple evaluation dimensions.
2Productivity
If traditional evaluation mechanism marks grouped objects as successfully identified, then the evaluation speed is improved, but the accuracy of scoring individual objects deteriorates
Solution Approach 1:
The patent performs preliminary action by pre-defining standard blocks for each individual object before evaluation. These standard blocks are stored and ready for comparison, allowing the system to quickly evaluate both grouped and individual object identification without sacrificing accuracy. The preliminary preparation enables fast evaluation while maintaining precise scoring.
Solution Approach 2:
The patent implements feedback by comparing prediction blocks against merged standard blocks and generating evaluation information that reflects both group-level and individual object identification accuracy. This feedback mechanism provides detailed scoring that maintains precision while enabling efficient evaluation through automated comparison and scoring algorithms.
3Ease of operation
If machine vision identifies multiple grouped objects as one successful identification, then the simplicity of evaluation is improved, but the reliability of efficiency measurement deteriorates
Solution Approach 1:
The patent applies dynamics by making the evaluation approach adaptive: the system can evaluate at different levels of granularity depending on the specific case. It dynamically switches between evaluating grouped objects as a whole and evaluating individual objects within groups, providing reliable efficiency measurement while maintaining operational flexibility and simplicity where applicable.
Solution Approach 2:
The patent resolves this contradiction by operating in multiple evaluation dimensions simultaneously. It maintains the simple group-level evaluation dimension while adding an individual object identification dimension through merged standard blocks. This multi-dimensional approach preserves simplicity for overall evaluation while ensuring reliability through detailed individual object scoring.
Data Source
AI summary
Embodiments of the present disclosure provide a method for evaluating (e.g., estimating) an efficiency of a machine vision, which includes: obtaining an image, wherein the image presents a plurality of objects which include a first object and a second object; performing an image recognition on the images by the machine vision to obtain a prediction block corresponding to at least one of the first object and the second object; merging a first standard block corresponding to the first object and a second standard block corresponding to the second object to obtain a third standard block; and obtaining evaluation information according to the third standard block and the prediction block, wherein the evaluation information reflects a prediction efficiency of the machine vision for the objects in the image.