TL;DR
This paper presents a new method for boresight calibration of aerial multi-head camera systems by leveraging the fixed geometric relationships between cameras as a constraint. This approach improves the accuracy and precision of estimating the crucial alignment angles between the camera and navigation systems, leading to more reliable direct georeferencing.
What Problem Does This Paper Address?
Aerial multi-head camera systems, composed of multiple cameras, are used to increase ground coverage and create a single synthetic image. To accurately georeference these images directly from GPS/INS navigation data, the precise alignment (boresight) between the camera and the Inertial Navigation System (INS) is critical. Small errors in these boresight angles can lead to significant inaccuracies in the final ground positioning. While INS and camera positions are generally well-known, precisely measuring the small rotational angles between their frames is challenging. This paper addresses the problem of accurately estimating these unknown boresight angles for multi-head camera systems.
Methods
The core of the paper's contribution lies in a novel geometric constraint applied to the boresight calibration process. The authors propose using the constant geometric calibration parameters between the individual cameras within the multi-head system as a constraint during bundle adjustment. This is based on the fact that these cameras are rigidly affixed to the platform, meaning their relative positions and orientations don't change.
The method involves:
- Estimating Camera Platform Geometric Calibration Parameters: These parameters define the fixed spatial relationship (positional displacements and rotation angles) between the master camera and each slave camera.
- Direct Georeferencing: Exterior Orientation Parameters (EOPs) for each camera are computed directly from onboard GPS/INS data and the camera platform geometric calibration parameters.
- Constrained Bundle Adjustment: A bundle adjustment is performed, where the collinearity equations for all cameras are simultaneously solved. The key innovation is incorporating the known camera platform geometric calibration parameters as fixed constraints within this adjustment. This helps to stabilize and improve the estimation of the unknown boresight angles.
Datasets
The paper describes a simulation testbed for their proposed method. They do not use real-world datasets for evaluation but rather simulate data based on realistic parameters:
- Target Area: Approximately 240m average ground height.
- Flight Parameters: Circular pattern, 30° oblique angle, 2500m radius, 5400m flying height.
- Ground Control Points (GCPs): 3 GCPs with ±1.0m precision.
- Tie Points: 41 tie points measured in overlapping areas.
- Simulated Errors: Random errors with ±2 pixel standard deviation added to image coordinates. Standard deviations for navigation solution positions and attitudes are ±2.0m and ±1.0°, respectively.
Key Insights and Ideas
- Leveraging Fixed Geometry: The most significant insight is that the fixed geometric relationships between cameras in a multi-head system are a powerful, previously underutilized constraint for boresight calibration.
- Constrained vs. Unconstrained: The fixed camera geometry constraint significantly improves the accuracy and precision of boresight estimation compared to methods that do not use this constraint (e.g., using only the master camera's data).
- Improved Direct Georeferencing: By more accurately estimating boresight angles, the overall accuracy of direct georeferencing for multi-head systems can be enhanced.
Key Results
The paper presents simulation results comparing a "1-camera case" (using only the master camera) with a "3-camera case" (using all three cameras with the proposed constraint).
- RMS Residuals: The 3-camera case showed better results (lower RMS residuals) in all X, Y, and Z directions compared to the 1-camera case.
- X-direction: 1-camera: ±0.3050m, 3-cameras: ±0.2384m
- Y-direction: 1-camera: ±0.2682m, 3-cameras: ±0.2438m
- Z-direction: 1-camera: ±0.4731m, 3-cameras: ±0.3442m
- Posteriori Variances of Unit Weight: The 3-camera case achieved a lower posteriori variance of unit weight (0.2708) compared to the 1-camera case (0.3419), indicating a better fit of the model to the observations.
Limitations and Future Work
- Simulation-Based: The results are based on simulated data. While the simulation parameters are realistic, real-world flight conditions and sensor noise might introduce additional complexities.
- Future Work: The authors suggest that this constrained method provides better estimation in terms of both accuracy and precision, implying that further validation with real flight data would be beneficial. The paper also mentions the importance of flight planning for calibration, which could be an area for future research.
Summary updated