Detection on Fallen People Dataset

Hey! We are releasing the detection of fallen people dataset. Stay tuned!

References

If you use the BIWI RGBD-ID dataset, please cite the following work (in preparation):

M. Munaro, A. Fossati, A. Basso, E. Menegatti and L. Van Gool.
"One-Shot Person Re-Identification with a Consumer Depth Camera, Book Chapter in "Person Re-Identification." 
Book Chapter in "Person Re-Identification", Springer, 2013.

BIWI RGBD-ID Dataset

The BIWI RGBD-ID Dataset is a RGB-D dataset of people targeted to long-term people re-identification from RGB-D cameras.

It contains 50 training and 56 testing sequences of 50 different people. The dataset includes synchronized RGB images (captured at the highest resolution possible with a Microsoft Kinect for Windows, i.e. 1280x960 pixels), depth images, persons' segmentation maps and skeletal data (as provided by Microsoft Kinect SDK), in addition to the ground plane coordinates. These videos have been acquired at about 10fps.

In the training videos, people performs a certain routine of motions in front of a Kinect, such as a rotation around the vertical axis, several head movements and two walks towards the camera.

28 people out of 50 present in the training set have been recorded also in two testing videos each. These testing sequences have been collected in a different day and in a different location with respect to the training dataset, therefore most subjects are dressed differently. For every person in the testing set, a Still sequence and a Walking sequence have been collected. In the Still video, every person is still or slightly moving in place, while in the Walking video, every person performs two walks frontally and two other walks diagonally with respect to the Kinect.

All the samples in the BIWI RGBD-ID Dataset are provided as folders with 5 different files for every frame:

• rgb image (1280x960 resolution)
• depth image (640x480 resolution)
• user map (640x480 resolution)
• txt file with skeleton tracker joint position and links orientation (estimated with Microsoft Kinect SDK)
• txt file with ground plane coefficients 

In the figure below, we report samples of rgb, depth, skeleton and user map data for five people of the training set:

Here below, samples from the testing sequences of the same people are shown:

For questions and remarks directly related to the BIWI RGBD-ID dataset please contact Questo indirizzo email è protetto dagli spambots. È necessario abilitare JavaScript per vederlo..

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. Copyright (c) 2013 Matteo Munaro.

RGB-D Re-Identification Datasets

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. Copyright (c) 2014 Matteo Munaro.

IAS-Lab RGBD-ID Dataset

The IAS-Lab RGBD-ID Dataset is a RGB-D dataset of people targeted to long-term people re-identification from RGB-D cameras.

It contains 11 training and 22 testing sequences of 11 different people. The dataset includes synchronized RGB images, depth images, persons' segmentation maps and skeletal data (as provided by OpenNI SDK), in addition to the ground plane coordinates. These videos have been acquired at about 30fps.

For every subject, we recorded three sequences, where the person rotates on himself and performs some walks. The first (Training) and the second (TestingA) sequences were acquired with people wearing different clothes, while the third one (TestingB) was collected in a different room, but with the same clothes as in the first sequence. These two different testing sets allow to validate both short-term and long-term re-identification techniques on this dataset.

Since NiTE skeletal tracking often poorly estimates the whole skeleton when some joints are not visible, we kept only those frames where all the joints are marked as tracked by the algorithm.

All the samples in the IAS-Lab RGBD-ID Dataset are provided as folders with 4 different files for every frame:

• rgb image (640x480 resolution)
• depth image (640x480 resolution)
• user map (640x480 resolution)
• txt file with skeleton tracker joint position and links orientation (estimated with NiTE middleware)

   

Moreover, a further txt file with ground plane coefficients is provided for every folder.

For questions and remarks directly related to the IAS-Lab RGBD-ID dataset please contact Questo indirizzo email è protetto dagli spambots. È necessario abilitare JavaScript per vederlo..

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. Copyright (c) 2014 Matteo Munaro.

CAVIAR4REID Skeleton Annotations

Further information will be added soon!

For a download link, please go to the 'Downloads' section.

For questions and remarks directly related to the CAVIAR4REID skeleton annotations please contact Questo indirizzo email è protetto dagli spambots. È necessario abilitare JavaScript per vederlo..

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. Copyright (c) 2014 Matteo Munaro.

IAS-Lab RGB-D Face Dataset

The IAS-Lab RGB-D Face Dataset has been created to measure accuracy and precision of 2D and 3D face recognition algorithms based on data coming from consumer RGB-D sensors. The Kinect v2 (or Kinect One) has been used to acquire this dataset.

The training dataset consists of 26 subjects captured in 13 different conditions (with pose, light and expression variations), standing 1 or 2 meters from the sensor.

In order to represent a typical service robotics scenario, where few people have to be recognized and many others have to be classified as unknown, the testing dataset contains 19 subjects and just four of them were also present in the training dataset. The other testing subjects can thus be considered as unknown.
The testing set is furthermore divided into five subsets, as explained in this table: 

The IAS-Lab RGB-D Face Dataset provides two different files for every frame:

• RGB image (1920x1080 resolution)
• XYZRGB point cloud (960x540 resolution)

The point cloud is registered to the RGB image and its resolution is downsampled by two. An example of the correspondence between RGB image and point cloud is available here.

The intrinsic parameters of the RGB camera are also provided together with the dataset in the "camera_info.yaml" file.

Samples from the training and the testing set are reported here below. The first four rows contain frames of the four known persons of the dataset, that are persons present in both the training and the testing dataset, while the other rows exemplify the "unknown" class. 

For obtaining the dataset, please refer to the Downloads section. 

For questions and remarks directly related to the IAS-Lab RGB-D Face dataset please contact Questo indirizzo email è protetto dagli spambots. È necessario abilitare JavaScript per vederlo..

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. Copyright (c) 2016 Matteo Munaro.

Downloads

BIWI RGBD-ID dataset download:

• the dataset as compressed rar files: training set (50 people), testing set (28 people)
• further information on the dataset and some Matlab files for reading and visualizing the dataset files are present in this zip folder. 

If you use the BIWI RGBD-ID dataset, please cite the following works:

M. Munaro, A. Fossati, A. Basso, E. Menegatti and L. Van Gool.
"One-Shot Person Re-Identification with a Consumer Depth Camera".
Book Chapter in "Person Re-Identification", pp 161-181, ISBN: 978-1-4471-6295-7, ISSN: 2191-6586, Springer, 2014.

M. Munaro, A. Basso, A. Fossati, L. Van Gool and E. Menegatti.
"3D Reconstruction of freely moving persons for re-identification with a depth sensor".
In Proceedings of IEEE International Conference on Robotics and Automation (ICRA), Hong Kong (China), pp. 4512-4519, 2014.

IAS-Lab RGBD-ID dataset download:

• the dataset as compressed rar files: Training, TestingA and TestingB.
• further information on the dataset and some Matlab files for reading and visualizing the dataset files are present in this zip folder. 

If you use the IAS-Lab RGBD-ID dataset, please cite the following works:

M. Munaro, S. Ghidoni, D. Tartaro Dizmen and E. Menegatti.
"A feature-based approach to people re-identification using skeleton keypoints".
In Proceedings of IEEE International Conference on Robotics and Automation (ICRA), Hong Kong (China), pp. 5644-5651, 2014.

M. Munaro, A. Basso, A. Fossati, L. Van Gool and E. Menegatti.
"3D Reconstruction of freely moving persons for re-identification with a depth sensor".
In Proceedings of IEEE International Conference on Robotics and Automation (ICRA), Hong Kong (China), pp. 4512-4519, 2014.

CAVIAR4REID skeleton annotations:

• the skeleton annotations as compressed rar files: annotations.

If you use our CAVIAR4REID skeleton annotations, please cite the following work:

M. Munaro, S. Ghidoni, D. Tartaro Dizmen and E. Menegatti.
"A feature-based approach to people re-identification using skeleton keypoints".
In Proceedings of IEEE International Conference on Robotics and Automation (ICRA), Hong Kong (China), pp. 5644-5651, 2014.

IAS-Lab RGB-D Face dataset download:

• the dataset as compressed zip files: training set (26 people), testing set (19 people)
• some C++ files for reading and visualizing the dataset files are present in this zip folder. 

If you use the IAS-Lab RGB-D Face dataset, please cite the following work:

G. Pitteri, M. Munaro and E. Menegatti.
"Depth-based frontal view generation for pose invariant face recognition with consumer RGB-D sensors".
In Proceedings of the 14th International Conference on Intelligent Autonomous Systems (IAS-14), Shanghai, 2016.

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. Copyright (c) 2016 Matteo Munaro.