Appendix - Developer Data Guide
Developer Data V1.1.0
1 Data Overview
| Data Type | Description | Target Path (output/developer_data/) |
|---|---|---|
| Raw images | Raw fisheye images captured by the Lixel device | raw/images |
| Human + device masks for raw images | Masks marking the positions of humans and devices | raw/masks |
| Camera extrinsics and intrinsics for raw images | COLMAP-format parameters aligned with the LAS point cloud | raw/sparse |
| Undistorted images | Undistorted perspective images | perspective/images |
| Human + device masks for undistorted images | Masks marking the positions of humans and devices | perspective/masks |
| Camera extrinsics and intrinsics for undistorted images | COLMAP-format parameters for perspective images aligned with the LAS point cloud | perspective/sparse |
| Optimized laser point cloud | Post-processed LAS point cloud | map_remov.las |
| Device poses | Device poses in the IMU coordinate system (10Hz) | poses.csv |
| High-frequency device poses | Device poses at the raw frequency (200Hz) | high_frequency_poses.csv |
2 Data Details
2.1 Raw Images, Masks, and Camera Parameters
Content: High-definition raw fisheye images captured by the Lixel device
Frame Rate: Adjusted according to the reconstruction quality option (low/medium/high)
Mask Details: Marks the pixel semantics in the image. Among them, 255 indicates regions unsuitable for reconstruction, such as humans, the capture device, etc.
Format: Provides the visual mapping results in COLMAP format aligned with the LAS point cloud, including the intrinsics and extrinsics of each image
Use Cases:
- Development of customized visual mapping algorithms
- Further optimization of existing visual mapping results
- Customized image undistortion operations based on specific business requirements
2.2 Undistorted Images, Masks, and Camera Parameters
Content: Ideal perspective images derived from raw fisheye images
Masks and Parameters: Includes the masks of the perspective images and the intrinsics and extrinsics in COLMAP format
Image Segmentation: Each fisheye image is segmented into 3 perspective images (60° wide, 90° high field of view). Two fisheye images can capture a panoramic image of horizontal 360° and vertical 90°
Use Cases:
- Suitable for 3DGS, MVS, or other algorithms that require ideal perspective images
2.3 Optimized Laser Point Cloud, Device Poses, and High-Frequency Device Poses
Content:
- Globally optimized LAS point cloud with dynamic point filtering capability
- Device poses in the Lixel IMU coordinate system (10Hz)
- High-frequency device poses at the raw IMU frequency (200Hz)
Coordinate Alignment:
- Without RTK: The LAS point cloud coordinates are in the same coordinate system as the extrinsics of the raw images and undistorted images
- With RTK: After the LAS point cloud coordinates are corrected by the offset in the LAS header, they are in the same coordinate system as the extrinsics of the raw images and undistorted images
Use Cases:
- Combined with raw or undistorted images for visual and laser post-processing reconstruction
Pose Data Format:
The pose data is provided in poses.csv and high_frequency_poses.csv files, in the following format:
timestamp, tx, ty, tz, qw, qx, qy, qz Where: 1. timestamp: time (seconds) 2. tx, ty, tz: translation from the IMU to the world coordinate system (meters) 3. qw, qx, qy, qz: rotation from the IMU to the world coordinate system (quaternion, unitless)