Feature Point Model Correction for Object Detection Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing technologies face challenges in precisely detecting objects with varying contour edge shapes due to individual differences and parallax effects, leading to suboptimal model pattern matching and detection accuracy.
Innovation Solution
An image processing device that detects objects by matching feature points from input data with a model pattern, using a corresponding point selection unit to associate second feature points with first feature points and a model pattern correction unit to calculate statistics for correcting the model pattern based on corresponding points, thereby improving detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a model pattern is created from a single reference object, then the model generation process is simple, but detection accuracy deteriorates when individual differences or parallax variations occur
Solution Approach 1:
The patent segments the model pattern creation process into multiple stages: first creating an initial model from one reference object, then iteratively refining it by incorporating data from multiple detected objects. This segmentation allows the system to start with simple model generation while progressively improving detection accuracy through accumulated learning from individual differences and parallax variations.
Solution Approach 2:
The patent performs preliminary actions by collecting and analyzing feature points from multiple detected objects before finalizing the model pattern. The system预先 (in advance) gathers data on individual differences and parallax effects, then uses this pre-collected information to correct and optimize the model pattern, ensuring high detection accuracy is achieved before actual detection operations begin.
2Measurement precision
If the model pattern is corrected based on accumulated detection data, then detection accuracy improves, but the processing time and computational complexity increase
Solution Approach 1:
The patent applies partial action by selectively correcting only those feature points that exhibit significant individual differences or parallax variations, rather than recalculating the entire model pattern. The system identifies and focuses on the most critical feature points that need adjustment, reducing computational overhead while maintaining detection accuracy.
Solution Approach 2:
The patent implements a feedback mechanism where detection results from multiple objects are fed back into the model pattern correction process. The system continuously monitors detection accuracy and uses this feedback to iteratively refine the model pattern, optimizing the balance between improved detection accuracy and processing time through adaptive learning.
3Ease of operation
If feature points are selected to represent object contours, then object shape can be detected, but individual differences in shape cause matching failures
Solution Approach 1:
The patent changes the parameters of feature points by adjusting their positions and characteristics based on statistical analysis of individual differences. The system modifies feature point parameters such as location, orientation, and scale to account for variations in object shapes, enabling reliable matching across different individuals while maintaining the ability to detect object contours.
Solution Approach 2:
The patent creates a composite model pattern that integrates feature points from multiple objects, combining their respective characteristics into a unified representation. This composite approach allows the model to capture common features across different individuals while accommodating individual variations, thereby improving matching reliability without losing the ability to detect diverse object shapes.
Data Source
AI summary
An image processing device that detects an image of an object from input data captured by a vision sensor, on the basis of a model pattern including a set of a plurality of first feature points representing the shape of the object, includes an object detection unit that detects images of the object by matching between a plurality of second feature points extracted from the input data and a plurality of first feature points forming the model pattern, a corresponding point selection unit that selects, for each image of the object, second feature points corresponding to the first feature points and stores the selected second feature points as corresponding points, and a model pattern correction unit that calculates a statistic of a predetermined physical quantity of the plurality of corresponding points associated with each first feature point and corrects the first feature point on the basis of the statistic.