Abstract
To reduce the optimization parameters in error alignment for the multi-arm mechanical tracking augmented reality (AR) system, an improved error alignment algorithm inspired by the Denavit-Hartenberg (D-H) convention is proposed. By introducing plane geometry theory to establish new coordinate systems, the researchers successfully reduced the number of calibration parameters from 22 to 14. Experimental results demonstrate that this algorithm simplifies optimization equations, shortens computing time, and effectively optimizes system errors.
Introduction
In mechanical AR tracking systems, a series of connected arms and joints utilize photoelectric encoders to track movement. Accurate registration—the precise overlay of digital information onto the real world—requires a high-fidelity transformation model from the base of the mechanical arm to the camera's coordinate system.
Traditional calibration methods often account for errors in all six degrees of freedom (x, y, z translations and rotations) for every joint. This approach typically results in a high-dimensional optimization problem (e.g., 22 parameters), which frequently leads to non-convergence or excessive computational latency, compromising the real-time AR experience.
Methods: The Plane Theory Approach
The proposed algorithm simplifies the mathematical model by leveraging "Plane Theory" to establish more rational coordinate systems. This method is based on three geometric assumptions:
- Axis-Plane Relationship: Each rotary encoder possesses a fixed axis, with rotation occurring in a plane perpendicular to that axis.
- Intersection as an Axis: The intersection of two such planes defines a line that serves as a shared axis for transition coordinate systems.
- Simplified Coordinates: Two axes of a joint's coordinate system are situated on the rotation plane, while the third aligns with the encoder axis.
By establishing coordinate systems only at plane intersections or along encoder axes, the model becomes significantly more streamlined. Instead of multiple rotation matrices per link to account for "twist" errors, the authors utilize a refined set of transformation matrices ($M_{12}, M_{23}, M_{34}, M_{4c}$) focusing only on necessary geometric deviations. The remaining 14 parameters are solved using nonlinear least-squares minimization (MATLAB's fmincon).
Experimental Setup
The algorithm was evaluated using a custom-built mechanical tracking tabletop AR system prototype. The hardware configuration includes photoelectric encoders for reading joint angles, a multi-arm mechanical structure, and a camera-LCD assembly for AR output, all interfaced via a USB data acquisition system.
Key Results and Discussion
The "Plane Theory" approach was compared against the previous standard method, yielding the following improvements:
| Metric | Previous Method | "Plane Theory" (Proposed) |
|---|---|---|
| Optimization Parameters | 22 | 14 |
| Number of Transformation Matrices | 20 | 13 |
The reduction in parameter space from 22 to 14 dimensions significantly enhanced the stability of the optimization process. Key findings include:
- Computational Efficiency: The reduced parameter count markedly shortened the time required for calibration.
- Convergence Reliability: The new algorithm consistently achieved stable solutions where the previous high-dimensional method often failed to converge.
- Registration Accuracy: The optimized parameters enabled accurate alignment of virtual elements on the physical display.
Conclusion
By adapting D-H convention principles through the lens of plane geometry, this research provides a robust solution for error alignment in multi-arm tracking systems. The reduction in optimization parameters not only speeds up computation but also ensures more reliable system convergence. Future work may explore the algorithm's performance under dynamic mechanical stress and its applicability to broader robotic calibration tasks.
Summary updated