Stereo-Cam
| Stereo-Camera Main | Platform | Calibration |
Stereo Vision Lidar IMU Dataset:
The Newer College Dataset revisits New College, Oxford with modern stereo camera and LIDAR sensors as well as precise ground truth.
Please cite:
The Newer College Dataset: Handheld LiDAR, Inertial and Vision with Ground Truth
Milad Ramezani, Yiduo Wang, Marco Camurri, David Wisth, Matias Mattamala and Maurice Fallon
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
November 2021
[Preprint] [PDF] [Video]
The primary dataset is provided in rosbag file format. We also provide a version divided into raw files (pngs, pcds, csv).
| n | name | description | video |
|---|---|---|---|
| 1 | short experiment | 0.5 loop of the quad Q, loop of the mid-section, loop of parkland, loop of quad, loop of park (1530 seconds) | video |
| 2 | long experiment | original dataset. 3.5 loops of quad, 2 loops of the mid-section and quad. 3 loops of parkland (2180 seconds) | video |
| 3 | prior map | accurate map of entire experimental area - 290 points. various other downsampled formats also available | video |
| 4 | calibration | calibration log files for use in Kalibr | none |
| 5 | quad with dynamics | 4 loops of quad with increasingly aggressive motion of device with swinging and fast walking (398 seconds) | video |
| 6 | dynamic spinning | aggressive spinning of the device while point at a corner of the quad. speeds of up to 3.5 rad/sec | video |
| 7 | parkland mount | some laps of the parkland, up the mound’s staircase, one loop around the mound (500 seconds) | video |
Video sample playlist
Videos showing the left camera images.
ROS Topics
Each file contains the following list of topics:
| Topic | Frequency [Hz] | Description |
|---|---|---|
/camera/accel/sample |
250 | RealSense accelerometer (Bosch BMI055) |
/camera/gyro/sample |
400 | RealSense gyroscope (Bosch BMI055) |
/camera/imu |
250 | RealSense IMU fused (see below) |
/camera/infra1/image_rect_raw |
30 | RealSense rectified left camera image |
/camera/infra2/image_rect_raw |
30 | RealSense rectified right camera image |
/os1_cloud_node/imu |
100 | Ouster OS-1 IMU (InvenSense ICM-20948) |
/os1_cloud_node/points |
10 | Point Cloud from Ouster OS-1 |
/os1_node/imu_packets |
100 | raw Ouster IMU data (see definition here) |
/os1_node/lidar_packets |
650 | raw Ouster LIDAR data (see definition here) |
/tf_static |
once | TF transforms between devices |
RealSense IMU interpolation
The RealSense SDK natively outputs two independent signals for gyro at 400 Hz and accelerometer at 250 Hz. Even though the driver provides some options to fuse the two messages together (either by copy or linear interpolation), we record the raw independent signals both for performance reasons and to give the dataset users the freedom to implement the interpolation of choice.
For those who need a unified message, we augmented the rosbag in post processing with the /camera/imu topic, which contains the accelerometer messages at 250 Hz
augmented with the linear interpolated values from the gyroscope. This gives a signal a nominal frequency of 250 Hz (same as accelerometer), rather than
a stream of messages unevenly spaced in time at 650 Hz on average.
Alternative Formats
We also provide the dataset as individual files within an organised folder system for use outside of ROS. This format matches that of the EuRoC dataset.
- Images are saved as PNG with images numbered according to their timestamp
- LIDAR data is saved in PCD pointclouds numbered according to their timestamp. These clouds are NOT motion corrected.
- IMU messages are saved as a csv file.