Dynamic Particle Range Adjustment for Attitude Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing attitude estimation methods using particle swarm optimization (PSO) face challenges in setting an appropriate occurrence range for random numbers, leading to particles being generated in excessively large ranges, hindering convergence to the actual attitude and increasing the risk of local solutions.
Innovation Solution
The proposed method sets a movable range for each joint based on the degree of noncoincidence between frames, narrowing the occurrence range of random numbers to match the actual attitude change, thereby improving the convergence speed and accuracy of attitude estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the occurrence range of random numbers is set to the maximum movable range of joints, then the particle generation covers the full possible attitude space, but the particles are generated in excessively large ranges hindering convergence to actual attitude
Solution Approach 1:
The patent applies dynamics by making the occurrence range of random numbers adjustable rather than fixed. The range is dynamically determined based on the degree of noncoincidence between consecutive images, allowing the search space to adapt to the actual motion magnitude. When motion is small, the range is narrowed to accelerate convergence; when motion is large, the range is expanded to ensure coverage.
Solution Approach 2:
The patent changes the parameter of random number occurrence range based on the degree of noncoincidence calculated from image differences. By using image noncoincidence as a feedback signal, the system adjusts the search space parameter adaptively, transforming a static parameter into a dynamic one that responds to actual motion conditions.
2Adaptability or versatility
If the occurrence range of random numbers is set to the maximum movable range, then the full attitude space is explored, but the risk of falling into local solutions increases
Solution Approach 1:
The system dynamically adjusts the search space range based on actual motion detected through image noncoincidence. This prevents the system from unnecessarily exploring the entire attitude space when only a small portion is actually needed, thereby reducing the risk of converging to local minima while maintaining sufficient exploration capability.
Solution Approach 2:
The patent introduces feedback by using the degree of noncoincidence between consecutive images to determine the occurrence range of random numbers. This feedback mechanism allows the system to learn from actual motion patterns and adjust its search strategy accordingly, improving reliability by avoiding unnecessary exploration of irrelevant regions.
3Ease of manufacture
If a fixed occurrence range is used for random numbers, then the method is simple to implement, but the convergence speed is slow and accuracy is reduced
Solution Approach 1:
The patent transforms the fixed occurrence range into a dynamic parameter that automatically adjusts based on image noncoincidence. This dynamic adjustment simplifies the overall system by eliminating the need for manual tuning of fixed ranges, while simultaneously significantly improving convergence speed and estimation accuracy.
Solution Approach 2:
The system performs self-adjustment by automatically determining the appropriate occurrence range based on the degree of noncoincidence calculated from image data. This self-service mechanism eliminates the need for external parameter tuning, maintaining implementation simplicity while achieving superior performance.
Data Source
AI summary
A method performed by a computer for attitude estimation includes: executing a first process that includes obtaining a first image; executing a second process that includes calculating a degree of noncoincidence between the obtained first image and a second image obtained before the first image; executing a third process that includes setting, from the degree of noncoincidence, and for each movable portion of a model as an attitude estimation target, a range where the movable portion is estimated to be movable in a frame of the first image; and executing a fourth process that includes outputting the range set for each movable portion as a generation range of a random number determining an amount of change of the each movable portion of the model to a generating process configured to generate a particle as a candidate for an attitude of the model in accordance with the generation range.