Abstract:We present a novel method for extrinsically calibrating a camera and a 2D Laser Rangefinder (LRF) whose beams are invisible from the camera image. We show that point-to-plane constraints from a single observation of a V-shaped calibration pattern composed of two non-coplanar triangles suffice to uniquely constrain the relative pose between two sensors. Next, we present an approach to obtain analytical solutions using point-to-plane constraints from single or multiple observations. Along the way, we also show that previous solutions, in contrast to our method, have inherent ambiguities and therefore must rely on a good initial estimate. Real and synthetic experiments validate our method and show that it achieves better accuracy than previous methods.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: how to achieve extrinsic calibration between a 2D laser range finder (LRF) and a camera from a single observation. Specifically, the paper proposes a new method. By using a V - shaped calibration pattern composed of two non - coplanar triangles, the relative pose between the two sensors can be uniquely determined with only one observation. This method solves the problems of inherent ambiguity and the need for a large amount of observation data in existing methods, and improves the accuracy and efficiency of calibration.
### Background of the Paper
In robotic systems, cameras and 2D laser range finders (LRF) are often used to calculate environmental geometric structures. 2D LRF is widely used because of its light weight, low cost and low power consumption. However, to utilize the measurement results of LRF and camera, the relative pose (i.e., extrinsic parameters) between them must be accurately known. This is a classic extrinsic calibration problem, and the goal is to determine the transformation relationship between two coordinate systems.
### Existing Problems
1. **Difficulty in Finding Corresponding Points**: The scanning plane of 2D LRF is invisible in a normal camera image, which makes it difficult to find corresponding points.
2. **Limitations of Existing Methods**:
- **Zhang and Price's Method**: It can only constrain two degrees of freedom and requires a large amount of observation data from different viewing angles.
- **Vasconcelos' Method**: Although a minimum solution is proposed, there are degenerate cases.
- **Zhou's Method**: An algebraic method is proposed, but multiple observation data are still required.
- **Other Methods**: Planar plates, triangular plates, V - shaped targets or rectangular plates with lines are used, but all have spatial ambiguity and require a large amount of observation data to reduce errors.
### Contributions of the Paper
1. **Unique Solution**: By minimizing the point - to - plane error, the calibration problem can be uniquely determined by observing two non - coplanar triangles once.
2. **Sufficient Constraints**: Existing methods fail to sufficiently constrain the calibration problem, resulting in the need for a large amount of measurement data and a good initial estimate. The method in this paper avoids these problems by adding geometric constraints.
3. **Analytical Solution**: A robust analytical solution for solving the point - to - plane constraint system from a single observation in the presence of noise is provided.
4. **High Precision**: Even if only a small amount of additional observation data is used, the method in this paper can significantly reduce errors and is superior to existing methods.
### Overview of the Method
1. **Geometric Constraints**: Through the V - shaped calibration pattern, three laser points and four plane normal vectors are defined, and point - to - plane constraint conditions are established.
2. **Analytical Solution**: Through nonlinear optimization and polynomial solving methods, the analytical solution of the relative pose is obtained.
3. **Robustness**: By excluding invalid solutions and optimization methods, the robustness of the algorithm is improved.
### Conclusion
The method proposed in this paper solves the inherent ambiguity problem of extrinsic calibration between 2D LRF and camera by introducing new geometric constraints, achieves a unique solution from a single observation, and shows higher accuracy and robustness in practical applications.