Object Recognition Integrated Score for Reliability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition methods face challenges in accurately determining the quality of position/orientation recognition results, leading to low recognition accuracy and unstable picking operations, especially when similarity scores have low reliability and are affected by illumination and shadows.
Innovation Solution
The proposed solution involves calculating an integrated score that combines similarity scores with reliability indices focusing on the diversity of the three-dimensional shape, allowing for improved recognition accuracy and stable picking operations by discriminating between high and low accuracy recognition results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a similarity score is used to determine the quality of position/orientation recognition results, then the recognition process is simple, but the recognition accuracy is low when the similarity score has low reliability
Solution Approach 1:
The patent combines multiple evaluation indices (similarity score, contour evaluation value, and reliability index) into a comprehensive evaluation system. The reliability index is calculated by integrating multiple factors including the diversity of three-dimensional shape, and the final recognition quality is determined by combining all these indices, thereby resolving the contradiction between simple process and accurate recognition.
Solution Approach 2:
The patent introduces a reliability index that changes based on the diversity of three-dimensional shape parameters. When the object has low three-dimensional shape diversity, the reliability index decreases, automatically adjusting the evaluation criteria to account for potential recognition errors, thus improving accuracy without overly complicating the process.
2Measurement precision
If contour evaluation value is used to improve recognition accuracy, then the accuracy may improve in some cases, but the evaluation becomes unreliable when affected by illumination and shadows
Solution Approach 1:
The patent introduces a reliability index as an intermediary evaluation metric that assesses the trustworthiness of the contour evaluation value. The reliability index is calculated based on the diversity of three-dimensional shape, and when illumination or shadows affect the contour detection, the low three-dimensional shape diversity results in a low reliability index, indicating that the contour evaluation should be weighted less or discarded.
Solution Approach 2:
The system uses the reliability index as feedback to adjust the weight or validity of the contour evaluation value in the final recognition result. When the reliability index is low (indicating potential issues with illumination or shadows), the system automatically reduces the influence of contour evaluation, thereby maintaining evaluation reliability.
3Stability of the object's composition
If recognition results with low accuracy are used for object picking, then the picking operation may be unstable, but re-measuring increases the time consumption
Solution Approach 1:
The patent performs a preliminary reliability assessment using the reliability index before finalizing the recognition result for object picking. By evaluating the diversity of three-dimensional shape and calculating the reliability index in advance, the system can identify potentially inaccurate recognition results before they are used for picking, allowing for timely re-measurement or alternative strategies without significant time loss.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A recognition result having low accuracy caused by a similarity score whose reliability is low is determined when recognizing a three-dimensional position/orientation of an object. An object recognition processing apparatus includes: a model data acquisition unit configured to acquire three-dimensional model data of an object; a measurement unit configured to acquire measurement data including three-dimensional position information of the object; a position/orientation recognition unit configured to recognize a position/orientation of the object based on the three-dimensional model data and the measurement data; a similarity score calculation unit configured to calculate a similarity score indicating a degree of similarity between the three-dimensional model data and the measurement data in a position/orientation recognition result of the object; a reliability calculation unit configured to calculate an index indicating a feature of a three-dimensional shape of the object, and calculate a reliability of the similarity score based on the index; and an integrated score calculation unit configured to calculate an integrated score indicating a quality of the position/orientation recognition result of the object based on the similarity score and the reliability.