研究目的
To develop a flexible and convenient calibration method for computing intrinsic and mechanical parameters of panoramic line-scan cameras using images of a simple calibration pattern, without unattainable assumptions.
研究成果
The proposed calibration method effectively computes intrinsic and mechanical parameters for panoramic line-scan cameras using a simple pattern and free poses. It avoids unattainable assumptions and provides closed-form solutions with iterative refinement. Experiments with synthetic and real data show accuracy comparable to standard calibration methods for classical cameras, with sub-pixel detection yielding reliable results. Future work should focus on degenerate configurations and integration into structure-from-motion frameworks.
研究不足
The method requires a sufficient number of images (at least 5-10) for accurate calibration. It may be sensitive to degenerate configurations, such as images taken from similar poses, which can lead to ill-conditioned systems. The calibration pattern must be designed to avoid curves too close to vertical directions, which can cause discontinuities. Real-time applications are limited due to the rolling-shutter effect on moving platforms.
1：Experimental Design and Method Selection:
The method uses a full geometrical model for panoramic line-scan cameras, involving intrinsic parameters (focal to pixel size ratio α and vertical principal point v0) and mechanism parameters (off-axis distance δ, and pitch, yaw, roll angles ψ, γ, φ). Calibration is performed in two steps: intrinsic calibration using 2D/3D point correspondences from a calibration pattern, and mechanical calibration using points at infinity derived from straight line projections. Closed-form solutions and iterative refinement are employed.
2：Sample Selection and Data Sources:
Synthetic data generated with a panoramic camera simulator (resolution 2048x6600 pixels) and real data from the PANORAMOS camera prototype. Calibration pattern consists of two planes with black/white stripes, roughly orthogonal, with known distances and angles.
3：List of Experimental Equipment and Materials:
PANORAMOS camera (CMOS type, gray level I/O, USB2, resolution 2048x6600 pixels, retina height 14.336 mm, lens focal length 6.5 mm), calibration pattern with specified properties (e.g., distance between lines r=2 cm, angle σ=10°).
4：336 mm, lens focal length 5 mm), calibration pattern with specified properties (e.g., distance between lines r=2 cm, angle σ=10°).
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: For intrinsic calibration, images of the pattern are taken from free poses. Points on the same image column are detected using edge detection. Cross-ratios are used to compute 3D coordinates. Intrinsic parameters are computed via Direct Linear Transformation and iterative refinement. For mechanical calibration, points at infinity are detected from projections of straight lines using parametric curve fitting. Mechanism parameters are computed linearly and refined iteratively.
5：Data Analysis Methods:
Singular Value Decomposition (SVD) for solving linear systems, Levenberg-Marquardt algorithm for nonlinear optimization, reprojection error minimization for validation. Synthetic data tests include noise addition (Gaussian noise with σ from 0.1 to 2 pixels) to assess robustness.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
