BelgiumTS Dataset    

Maintained by Radu Timofte (email: radu.timofte at

Envisaged structure of BelgiumTS:
BelgiumTS/structure.txt   2747 bytes
Defined Traffic Signs

BelgiumTS/DefinedTS.tar.gz   1398198 bytes
BelgiumTS/reducedSetTS.txt   281 bytes -- reduced set of 62 traffic sign types as used in the experiments from [1,2]


The cameras are taken in clockwise order 0-7, are mounted in pairs, one pair on each side of the recording van. 00 is the frontal left and 01 is the frontal right camera.

BelgiumTS/Annotations/camera00.tar   220897280 bytes
BelgiumTS/Annotations/camera01.tar   948019200 bytes
BelgiumTS/Annotations/camera02.tar   377630720 bytes
BelgiumTS/Annotations/camera03.tar   28149760 bytes
BelgiumTS/Annotations/camera04.tar   72325120 bytes
BelgiumTS/Annotations/camera05.tar   659650560 bytes
BelgiumTS/Annotations/camera06.tar   174295040 bytes
BelgiumTS/Annotations/camera07.tar   244193280 bytes
BelgiumTS/Annotations/annotations.tar   1822720 bytes

BelgiumTS/Annotations/   485645 bytes (provided for detection evaluation, it uses the same material and splits)

BelgiumTS for Classification (cropped images): BelgiumTSC_Training (171.3MBytes) BelgiumTSC_Testing (76.5MBytes)

Background images:

BelgiumTS/NonTSImages/NonTS_TestingBG.tar   176414720 bytes
BelgiumTS/NonTSImages/NonTS_TrainingBG.tar00   1047527424 bytes
BelgiumTS/NonTSImages/NonTS_TrainingBG.tar01   1047527424 bytes
BelgiumTS/NonTSImages/NonTS_TrainingBG.tar02   1047527424 bytes
BelgiumTS/NonTSImages/NonTS_TrainingBG.tar03   1047527424 bytes
BelgiumTS/NonTSImages/NonTS_TrainingBG.tar04   664254464 bytes

Test sequences:
Cameras, poses, annotations   17196942 bytes

Sequence 01 - 3001 frames x 8 cameras
BelgiumTS/Seqs/Seq01_00.tar   908072960 bytes
BelgiumTS/Seqs/Seq01_01.tar   908072960 bytes
BelgiumTS/Seqs/Seq01_02.tar   908062720 bytes
BelgiumTS/Seqs/Seq01_03.tar   908072960 bytes
BelgiumTS/Seqs/Seq01_04.tar   908062720 bytes
BelgiumTS/Seqs/Seq01_05.tar   908062720 bytes
BelgiumTS/Seqs/Seq01_06.tar   908072960 bytes
BelgiumTS/Seqs/Seq01_07.tar   908072960 bytes

Sequence 02 - 6001 frames x 8 cameras
BelgiumTS/Seqs/Seq02_00.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq02_00.tar01   828819456 bytes
BelgiumTS/Seqs/Seq02_01.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq02_01.tar01   828819456 bytes
BelgiumTS/Seqs/Seq02_02.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq02_02.tar01   828798976 bytes
BelgiumTS/Seqs/Seq02_03.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq02_03.tar01   828798976 bytes
BelgiumTS/Seqs/Seq02_04.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq02_04.tar01   828809216 bytes
BelgiumTS/Seqs/Seq02_05.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq02_05.tar01   828809216 bytes
BelgiumTS/Seqs/Seq02_06.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq02_06.tar01   828829696 bytes
BelgiumTS/Seqs/Seq02_07.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq02_07.tar01   828809216 bytes

Sequence 03 - 2001 frames x 8 cameras
BelgiumTS/Seqs/Seq03_00.tar   605491200 bytes
BelgiumTS/Seqs/Seq03_01.tar   605480960 bytes
BelgiumTS/Seqs/Seq03_02.tar   605480960 bytes
BelgiumTS/Seqs/Seq03_03.tar   605480960 bytes
BelgiumTS/Seqs/Seq03_04.tar   605480960 bytes
BelgiumTS/Seqs/Seq03_05.tar   605480960 bytes
BelgiumTS/Seqs/Seq03_06.tar   605480960 bytes
BelgiumTS/Seqs/Seq03_07.tar   605480960 bytes

Sequence 04 - 4001 frames x 8 cameras
BelgiumTS/Seqs/Seq04_00.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq04_00.tar01   163147776 bytes
BelgiumTS/Seqs/Seq04_01.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq04_01.tar01   163137536 bytes
BelgiumTS/Seqs/Seq04_02.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq04_02.tar01   163065856 bytes
BelgiumTS/Seqs/Seq04_03.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq04_03.tar01   163055616 bytes
BelgiumTS/Seqs/Seq04_04.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq04_04.tar01   163117056 bytes
BelgiumTS/Seqs/Seq04_05.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq04_05.tar01   163127296 bytes
BelgiumTS/Seqs/Seq04_06.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq04_06.tar01   163127296 bytes
BelgiumTS/Seqs/Seq04_07.tar00   1047527424 bytes
BelgiumTS/Seqs/Seq04_07.tar01   163127296 bytes

Each time the data was >= 1 Gbyte we split the tar archive as in the example:
tar -cf - NonTSImages/TrainingBG | split -d -b 999m - NonTS_TrainingBG.tar

To untar you may use:
tar -xf archive.tar
cat NonTS_TrainingBG.tar* | tar -x

The dataset is hosted by ETH Zurich, [].
The older address [] is no longer available.

Radu Timofte moved to Computer Vision Lab, D-ITET, ETH Zurich
Home page :
Email : radu.timofte at

VISICS team wins the The German Traffic Sign Detection Benchmark with perfect results for Prohibitory and Danger traffic signs and top results for Mandatory signs.
Solution detailed in our IJCNN 2013 paper[pdf]

Radu.Timofte@VISICS is the 3rd scoring team in the German Traffic Sign Recognition Challenge (an IJCNN2011 competition)

Dataset released and hosted by VISICS, ESAT, KU Leuven


[!!!] If you use the material from this dataset, please cite one of the following publications, accordingly:

[6] Radu Timofte*, Markus Mathias*, Rodrigo Benenson, and Luc Van Gool, Traffic Sign Recognition - How far are we from the solution?, International Joint Conference on Neural Networks (IJCNN 2013), August 2013, Dallas, USA. (* equal contributions) [pdf][bib]
-> (BTSD and BTSC dataset splits for 2D TSR)

[5] Radu Timofte, Karel Zimmermann, and Luc van Gool, Multi-view traffic sign detection, recognition, and 3D localisation, Journal of Machine Vision and Applications (MVA 2011), DOI 10.1007/s00138-011-0391-3, December 2011, Springer-Verlag. [pdf][bib]
-> (introduces BelgiumTS dataset, BTSC and BTSD are using the BTS material and splits)

[4] Radu Timofte and Luc Van Gool, Sparse Representation Based Projections. In British Machine Vision Conference (BMVC 2011), 2011, UK. [pdf] [bib]
-> (introduces the BTSC part for classification, using the BTS material and splits)

[3] Radu Timofte, Victor Adrian Prisacariu, Luc J. Van Gool, and Ian Reid, Combining Traffic Sign Detection with 3D Tracking Towards Better Driver Assistance. Emerging Topics in Computer Vision and its Applications (editor: C.H. Chen). World Scientific Publishing. September 2011. [pdf][bib]
-> (uses the BTSC part for classification, and BTS for detection and tracking)

[2] Victor Adrian Prisacariu, Radu Timofte, Karel Zimmermann, Ian Reid, Luc van Gool, Integrating object detection with 3D tracking towards a better driver assistance system, IAPR International Conference on Pattern Recognition, ICPR 2010. [pdf][bib].
-> (uses the BTS for detection, classification, and tracking)

[1] Radu Timofte, Karel Zimmermann, Luc van Gool, Multi-view traffic sign detection, recognition, and 3D localisation, IEEE Workshop on Applications of Computer Vision, WACV 2009. [pdf][bib]
-> (uses the BTS for detection, classification, and 3D mapping of traffic signs)

The authors acknowledge to GeoAutomation for providing the data.