PIROPO Database
News
Published: C.R. del Blanco, P. Carballeira, F. Jaureguizar, N. García, “Robust people indoor localization with omnidirectional cameras using a Grid of Spatial-Aware Classifiers”, Signal Processing: Image Communication, vol. 93, article 116135, 2021. (doi: https://doi.org/10.1016/j.image.2021.116135)
Added calibration information for the omnidirecional camera in Room B. The file "calib_cameras.zip" contains now all the calibration data for the omnidirectional cameras.
Added calibration information for the omnidirecional cameras in Room A, and the floor plan. See section "How to get the PIROPO database" below.
About the database
The PIROPO database (People in Indoor ROoms with Perspective and Omnidirectional cameras) comprises multiple sequences recorded in two different indoor rooms, using both omnidirectional and perspective cameras. The sequences contain people in a variety of situations, including people walking, standing, and sitting. Both annotated and non-annotated sequences are provided, where ground truth is point-based (each person in the scene is represented by the point located in the center of its head). In total, more than 100,000 annotated frames are available.
Rooms, cameras, and video sequences
Room A is a rectangular-shaped open area, with an approximate size of 15 x 10 meters, and it is covered by 8 different cameras. Three of the cameras are omnidirectional (omni_1A, omni_2A, and omni_3A), and are ceiling-mounted. The other five cameras are perspective ones (conv_4A to conv_8A), and are installed in wall-ceiling edges. The image dimensions are 800 x 600 pixels for all cameras, and the frame rate is approximately 10 fps.
Room B is a smaller rectangular-shaped room of 8 x 5 meters, with a more cluttered background including multiple tables, chairs, and computers. Two cameras installed in the room have been used. One of them is omnidirectional (omni_1B), and it is ceiling-mounted, and the other is a perspective one (conv_2B). The image dimensions are 800 x 600 pixels for the omnidirectional camera, and 704 x 576 for the perspective one. The frame rate is approximately 10 fps.
Sample images of the cameras captured scenes can be found here. In both rooms, all the cameras have been used to record the same sequences. Therefore, there are as many sequences as number of cameras per recording case. However, notice that the sequences are not synchronized.
Description of the sequences
Several sequences have been captured, showing people in a variety of situations. The following table provides a brief description of the contents all the sequences. Sample videos showing the sequences for two of the database cameras can be found here.
Ground Truth Annotations
The PIROPO database includes ground truth annotations for some of the sequences (see table below). The annotations are contained in Comma Separated Values (.csv) files. Each row of the file represents one image of the corresponding sequence. Then, the first column indicates the number of the image within the sequence (1-indexed). The following column are organized in pairs, indicating the x-coordinate (image column) and the y-coordinate (image row) of the point belonging to the center of the head of each person in scene. Points (0,0) represent non-annotations for convenience purposes.
· Example of .csv annotation file contents for a sequence with at most one person in scene:
119,0,0 //frame 119; empty frame
120,0,0 //frame 120; empty frame
121,718,241 //frame 121; person (head) at x = 718 pixels, y = 241 pixels
· Example of .csv annotation file contents for a sequence with more than one person in scene:
11225,225,141,0,0,0,0 //frame 11225; person #1 at x = 225, y = 141; no more people in scene
11226,213,130,53,564,0,0 //frame 11226; person #1 at x = 213, y = 130; person #2 at x = 53, y = 564
(two complete .csv sample files can be found here and here)
· Available sequences and ground truth (GT) annotations in the database.- some sequences are not available for certain cameras, since the action takes place out of their field of view. Also, some sequences have not been annotated due to time restrictions. The following table provides information about the availability of sequences and annotations.
How to get the PIROPO database
The whole set of video sequences and associated data (ground truth) can be downloaded from the following two links (Room A and Room B):
Citations and rights
If you publish any work reporting results using this dataset, please cite the following paper:
C.R. del Blanco, P. Carballeira, F. Jaureguizar, N. García, “Robust people indoor localization with omnidirectional cameras using a Grid of Spatial-Aware Classifiers”, Signal Processing: Image Communication, vol. 93, article 116135, 2021. (doi: https://doi.org/10.1016/j.image.2021.116135)
The PIROPO database is available and free for research purposes. Contact cda@gti.ssr.upm.es for more information.