Different benchmark datasets are provided. Each single dataset includes:

• three image sequences, each sequence captured with a fisheye camera
• the laserscanner data
• an interactive 3D point cloud of the laserscanner data
• the data conventions
• a video of the scene to get first impressions
• a link to download the complete test data

Data Acquisition

All datasets are recorded with the same equipment.

The setup for data acquisition includes the helmet system, a high performance laptop, the tracking system and PoE cables. The tracking system is essential for tracking the trajectory of the helmet system as mentioned above. The tracking system use 8 cameras which have to be well placed for the different scenarios. The helmet system is connected with a laptop for saving the data obtained by the laserscanner and MCS.

Firstly the program Motive:Tracker is started on the laptop and the tracking system is calibrated with a calibration bar. Secondly the recording starts. Therefore an operator is wearing the helmet system on his head and moves to the different areas of interest. The main program is started with the connect button to capture the data.

Results

Then the laserscanner data and the images are obtained. Further from all scenes a video is recorded with an external digital camera to get a first impression.

Five benchmark datasets are available to download. The following table gives an overview of the provided datasets with the name of the dataset, the duration, the number frames per camera and the number of lines from the laser scanner, as well as the number of points of the point cloud.

Output

The provided data for the laserscanner is the scanning angle, the distance and the intensity for each 3D point. To maintain a constant number of measurements, only the first laser return and first intensity are used. Then the final number of the laserscanner measurements is 1080 per laserscanner rotation.

Provided ASCII .txt-files:

• LS_Dist.txt
• LS_Dir.txt
• LS_Intensity.txt
• Lspoint.txt
• Quaternion.txt
• Tracker.txt

and all the images (.png).

LS_Dist.txt contains the distances between the laserscanner and illuminated surface:
Header of the file: Internal-Timestamp* | Sensor-Timestamp** | Distances
The distances are in mm.

LS_Dir.txt contains the direction of the points:
Header of the file: Internal-Timestamp* | Sensor-Timestamp** | Directions
The angles are in degree.

LS_Intensity.txt contains the intensity of each point:
Header of the file: Internal-Timestamp* | Sensor-Timestamp** | Intensities

Lspoint.txt contains the coordinates of the points in the laserscanner coordinate system:
Header of the file: Timestamps* | Lspoint
The coordinates are in m.

Quaternion.txt contains the position of the sensor in Quaternion representation:
Header of the file: Internal-Timestamp* | (Sensor-Position) | (Quaternion: X Y Z W)
The coordinates are in m.

Tracker.txt contains the position of the marker 2 of the rigid body:
Header of the file: Internal-Timestamp* | NatNet-Timestamp | (Sensor-Position) | (Sensor-Coordinate-System-X-Axis) | (Sensor-Coordinate-System-Y-Axis) | (Sensor-Coordinate-System-Z-Axis)
The coordinates are in m.

* Internal-Timestamp: The general time stamp, which can be synchronized with the other sensors.
** Sensor-Timestamp: The time stamp of the sensor, which cannot be used for synchronization.

The images are provided with a coded filename: Timestamp_Camera number_img_number of pictures.png
For example: 6562617_Cam1_img_10000756.png

A large set of data scenes are acquired. Part of them are recorded indoor in a laboratory. In the laboratory it is a typical office environment with a size of ####x#### m². It includes tables, chairs, computers and a lot of clutter. In the video of the scene the room decomposition is easier to visualize. As in a basement, there is very little daylight. Besides the window blinds that are closed and the overhead light that was turned on. The setup of the tracking system in the laboratory causes no problems because the cameras of the tracking system are fixed to the wall.

For the dataset indoor_dynamic the mentioned laboratory room is used. The operator is moving in loops around the room and people are strolling around him.  The scene is recorded to test the ability of the tracking system to measure with people that are temporary recovering the helmet system.

The duration is 56 seconds with 899 frames per camera of the MCS. The laser scanner generates 2333 lines. With 1080 points per laser scan, it results in a point cloud of 2.52 million points.

Download the indoor_dynamic Dataset

For the dataset Indoor_static the laboratory room is used again, but this time without people strolling around. It is a static scene, where the operator is moving around the room as a loop and then moving up and down. The scene is recorded to see the difference between a dynamic scene with people strolling around and a static scene without people.

The duration is 63 seconds with 1014 frames per camera of the MCS. The laser scanner generates 2609 lines. It results in a point cloud of 2.52 million points.

Download the indoor_static Dataset

This scenario is outside in an atrium with a size of ####x#### m². The atrium is surrounded by facades. Only a part of the atrium is used for the test scenes. This part is on the edge of the atrium without high structures visible inside of the scene. The scene is composed of plants and small concrete blocks. As it is outside daylight is present. This can create problems if the sun is shining because the spherical rigid body markers are less visible for the tracking system. On the day of the recordings it was mostly cloudy.  

The following point cloud gives an impression of the scenario:

Download of the high resolution images which are used to calculate the point cloud. (4.4G)

For this scenario the operator is rotating around his vertical axis (static scene). The textured 3D point cloud derived by the laser scanner is shown below. The dataset is recorded for comparing it to the results of a terrestrial laser scanner.

The duration is 52 seconds with 779 frames per and the laser scanner generates 26173lines. It results in a point cloud of 2.35 million points.

Download the outdoor_rotation Dataset

The operator is moving in an outdoor environment (static scene). The movement is not a loop, the operator is strolling from the back to the front and from the left to the right in lines. As the trajectory is not a closed loop, no drift can be calculated.

Download the outdoor_static Dataset

The operator is moving in an outdoor environment (static scene). The movement is not a loop, the operator is strolling arround randomly. As the trajectory is not a closed loop, no drift can be calculated.

Download the outdoor_static2 Dataset

For this scene the operator is moving out of the range of the tracking system and is moving around the whole atrium in a closed loop. This scene is taken to test if the system is able to hold the trajectory without connection to the tracking system. It is ensured that the scene contains a closed loop to allow to recognize previously visited areas and afterwards to determinate and reduce the drift of the trajectory.

The duration is 225 seconds with 3176 frames per and the laser scanner generates 7993 lines. It results in a point cloud of 8.64 million points.

Download the outdoor large loop Dataset