This page gives access to the dataset of the acquisition campaign of soccer data during the F.I.G.C.- C.N.R. project.
Manually annotated objects position of two minutes of the
game.
These metadata provide the positions of the players, referees, and
ball in each frame of each camera (6 XML files).
The players have the same labels while they move in the six views that
correspond to the numbers on their uniforms. The player labels of the first team start from 1 while the player labels of the second teams start from 201.
The first 300 frames of each sequence have not been labelled in order to
provide an initial phase to initialize the background subtraction algorithms.
COMPOSITION
- Sequences
Six synchronized views acquired by six Full-HD cameras, three for each
major side of the playing-field, at 25 fps (6 AVI Files, about 300MB)
DOWNLOAD THE SEQUENCE
- Calibration Data
These are the pictures containing some reference point in to the
playing-field and the relative measures for calibrating each camera into a common world coordinate system (a zip file containing 6 pdf files, and 6 .bmp images)
DOWNLOAD THE ANNOTATION FILES
Terms of use
This dataset is available for non-commercial research in video signal processing only. We kindly ask you to
cite:
“T. D’Orazio, M.Leo, N. Mosca, P.Spagnolo, P.L.Mazzeo
A Semi-Automatic System for Ground Truth Generation of Soccer Video Sequences
6th IEEE International Conference on Advanced Video and Signal Surveillance, Genoa, Italy September 2-4 2009”
when using this dataset
Marine Protected Area
The dataset has been acquired in the MPA of Porto Cesareo - Lecce - Italy.
The sequences can be used for testing several IP algorithms:
- segmentation in challenging context;
- people counting;
- people and boat tracking.
Sequences will be available soon
This dataset is finalized to boat re-identification.
It is composed by 5523 images of 107 boats.
About 800 images have been acquired by a Axis Q6035-E Network Camera; other images have been obtained by applying* the following
transformations:
- Rotation
- Flipping
- Noise addiction
- Contrast
- Pixel level Inversion
- Blurring
Labelling rules - consider the file named ”0_c1_00_14.jpg” as example.
- ”0” stands for the boat ID;
- ”c1” represents the camera ID;
- "00" stands for original image, while ”000”,”001”,”002”,”003”,”004” and ”005” indicates the applied transformation in the same order as listed below ("000" is rotation, "001" is flipping, ...);
- The last field is only used to assign a progressive number to images.
*ImgAug library has been used for this - https://github.com/aleju/imgaug
Terms of use
This dataset is available for non-commercial research in video signal processing only. We kindly ask you to
cite:
“P. Spagnolo, F. Filieri, C. Distante, P.L. Mazzeo, P. D'Ambrosio, A new annotated dataset for boat detection and re-identification 16th IEEE International
Conference on Advanced Video and Signal Surveillance, Taipei, Taiwan Sept. 18-21 2019”
when
using this dataset