# NEGS-UGV Dataset



Paper:

## Overview

The dataset is classified in different folders by scenario and sequence. 
Each sequence has a name corresponding to its content. Inside each sequence 
there are different directories for each "device" (i.e. rgb image, imu, ...).

The different available devices are:

* RGB realistic stereo camera.

* RGB tagged stereo camera.

* Inertial Measurement Unit (IMU).

* LiDAR OS1.

* Global Navigation Satellite System (GNSS).

* Wheel tachometers.

* Pose Ground Truth.


In every case the timestamp is the ROS time recorded in the original rosbag file for that message.

## Images

The dataset contains two sets of stereo images. One corresponds to the RGB spectrum and the other is the label information.
The name convention followed is:

    timestamp_camera.jpg
and for the labeled one:
    timestamp_camera_tag.jpg

The images are stored in different directories named:

    img_right/

    img_right_tag/

    img_left/

    img_left_atg/

The capture ratio is 25 fps for both cameras. The image sizes are 1280x720 pixels for both cameras.  

## IMU

The data from the IMU is stored in a single "imu_data.txt" file. The file stores the messages as successive lines.
The structure for each line is:

    timestamp quaternion.x quaternion.y quaternion.z ang_vel_X ang_vel_Y ang_vel_Z lin_acc_X lin_acc_Y lin_acc_Z

- Roll, pitch and yaw are in rad and about fixed axes XYZ.

- Angular velocities (ang_vel_{X, Y, Z}) are in rad/sec.

- Linear accelerations (lin_acc_{X, Y, Z}) are in m/s².

The capture ratio is 50 .

## OS1 points

Each message is stored in one .csv file as a Nx4 matrix (being N the number of points per message) following the next convention:

    x y z intensity

- x, y and z are the position respect the Velodyne sensor in meters.

- The intensity correponds to one of the 15 classes.

-

They are stored in the folder `lidar/`. The name convention followed is:

    timestamp.csv

In order to open and parse the data from the binary file, the followings commands are examples of use:

* `numpy:`

```python
    import numpy as np
    
    points = np.fromfile(filename, dtype=np.float32).reshape(-1, 4)
```

* `MATLAB:`

```matlab
    fileID = fopen(filename, 'r');
    format = 'float32';
    data = fread(fileID, Inf, format);
    xyzi=reshape(data,4,length(data)/4)';
```

## GNSS

The data from the GNSS are stored in a single "GNSS_data.txt" file. The file stores the messages as successive lines.
The structure for each line is:

    timestamp latitude longitude altitude

## Wheel tachometers

The tachometer readings of each wheel are provided in four separated text files. Each line contains the timestamp and the wheel speed  in rad/s:

    timestamp wheel_speed

## Pose ground truth
The data from the exact positon of the "base_link" of the robot in the Gazebo globlal reference system. The file stores the messages as successive lines.
The structure for each line is:

    timestamp position_x position_y position_z orientation_x orientation_y orientation_z orientation_w
 
The dataset description and tools to work with it are available at: 
https://www.uma.es/robotics-and-mechatronics/info/132852/negs-ugv-dataset