Abstract
A spherical camera can observe the environment for almost 720 degrees’ field of view in one shoot, which is useful for augmented reality, environment documentation, or mobile mapping applications. This paper aims to develop a spherical photogrammetry imaging system for the purpose of 3D measurement through a backpacked mobile mapping system (MMS). The used equipment contains a Ladybug-5 spherical camera, a tactical grade positioning and orientation system (POS), i.e. SPAN-CPT, and an odometer, etc. This research aims to directly apply photogrammetric space intersection technique for 3D mapping from a spherical image stereo-pair. For this purpose, several systematic calibration procedures are required, including lens distortion calibration, relative orientation calibration, boresight calibration for direct georeferencing, and spherical image calibration. The lens distortion is serious on the ladybug-5 camera’s original 6 images. Meanwhile, for spherical image mosaicking from these original 6 images, we propose the use of their relative orientation and correct their lens distortion at the same time. However, the constructed spherical image still contains systematic error, which will reduce the 3D measurement accuracy. Later for direct georeferencing purpose, we need to establish a ground control field for boresight/lever-arm calibration. Then, we can apply the calibrated parameters to obtain the exterior orientation parameters (EOPs) of all spherical images. In the end, the 3D positioning accuracy after space intersection will be evaluated, including EOPs obtained by structure from motion method.
Introduction
Spherical cameras, such as the Ladybug-5, are increasingly popular for capturing immersive environments in a single exposure. While widely used in applications like Google Street View, their utility for high-precision 3D photogrammetry is often limited by complex geometric distortions and stitching artifacts. This paper presents a systematic calibration workflow for a backpack-mounted mobile mapping system (MMS) designed to transform spherical imagery into a reliable tool for centimeter-level 3D measurement.
System Design and Calibration Challenges
The hardware integration includes a Ladybug-5 camera, a SPAN-CPT tactical-grade Positioning and Orientation System (POS), and an odometer. Traditional spherical stitching software often relies on feature-based matching, which can fail in textureless environments (e.g., clear skies or indoor hallways) and introduces parallax errors if the object distance is not correctly estimated. To achieve professional-grade mapping accuracy, the authors developed a pipeline that moves away from pixel-matching toward a rigorous geometric calibration of the system's internal sensors.
Methodology
The proposed workflow involves a multi-stage calibration process to ensure geometric integrity:
- Interior Orientation and Lens Distortion: Each of the six individual CCD sensors within the Ladybug-5 is calibrated using an indoor field with over 200 coded targets to solve for focal length, principal points, and radial distortion.
- Relative Orientation (ROP): The exact spatial relationship (rotation and offset) between the "master" camera and the five "slave" cameras is determined.
- Geometric Stitching: Instead of matching features, the original images are projected onto a virtual sphere using the calibrated ROPs. This ensures the stitching is independent of object distance and eliminates "ghosting" artifacts.
- Direct Georeferencing (DG): For outdoor applications, the "lever-arm" (physical offset between the GNSS antenna and camera) and "boresight" (rotation between the IMU and camera) are calibrated to align the system with global coordinates.
Results and Discussion
The system was evaluated using two methods for determining camera position: Structure from Motion (SfM), which calculates position from image points, and Direct Georeferencing (DG), which relies on the onboard GNSS/IMU.
- Indoor Accuracy (SfM): The system achieved a 3D positioning accuracy of 5.6 cm, which is highly suitable for indoor mapping. The image space error was approximately 2.79 pixels.
- Outdoor Accuracy (DG): Performance was significantly lower outdoors, with mean errors reaching several meters. This is attributed to the inherent systematic errors of the multi-camera construction and the high-frequency vibrations of human gait, which the current POS Kalman filter struggles to process.
Conclusion
This research demonstrates that a backpacked spherical imaging system can achieve high-precision results through rigorous systematic calibration, particularly in indoor environments using SfM. Future work will focus on refining stitching algorithms to correct sensor-specific systematic errors and developing POS solutions better optimized for the unique dynamics of a human carrier to improve outdoor direct georeferencing accuracy.
Summary updated