The experiments for the Make and Model Vehicle Recognition (MMVR) system use
a Training Database (ISIR_LPReditor_dbtrain) along with a Test Database
(ISIR_LPReditor_dbtest).
These databases contain information of 20 different vehicle types, where
each type was placed in a specific folder:
- The Training Database comprises 166 high quality frontal vehicle images
in different scenarios: streets, outdoor and indoor car parks.
Below are shown a few examples:
- The Test Database contains 708 outdoor frontal vehicle images in different
light conditions and with lower resolution. The images where recorded using two
video camcorders with an optical zoom.
The following table displays the databases' composition:
Class | Test | Train |
Ci_Berlingo_A | 14 | 1 |
Ci_C3 | 41 | 1 |
Ci_Picasso_B | 28 | 4 |
Ci_Saxo_B | 23 | 9 |
Fi_Punto_A | 22 | 3 |
Pe_206_B | 51 | 10 |
Pe_307 | 43 | 12 |
Pe_405 | 24 | 6 |
Pe_406_B | 30 | 6 |
Re_19_B | 36 | 5 |
Re_Clio_A | 29 | 7 |
Re_Clio_C | 47 | 15 |
Re_Clio_D | 80 | 31 |
Re_Laguna_B | 34 | 5 |
Re_Scenic_B | 25 | 4 |
Re_Scenic_C | 27 | 4 |
Re_Twingo_A | 50 | 19 |
Re_Twingo_B | 34 | 15 |
VW_Golf_C | 40 | 4 |
VW_Golf_D | 30 | 5 |
Databases' labels
The type folders each contain a "fts" folder with a data file for every single
vehicle image.
Each type folder contains a "fts" folder with a data file for every single vehicle image.
img.jpg -> fts\img_plq.dat
The data files have information of the four-corners coordinates of the vehicle's
license plate.
The information in the
img_plq.dat file is the line :
540 666 820 661 824 715 540 719
providing (x,y) coordinates of the four points: A = (540,666), B = (820,661), and so on.
A ROI is defined by the points {A,B,C,D}, being independent of the location of
the vehicle in the image or of its scale. In order to correct the orientation
of the original image, an affine transformation allows to move original points
{A,B,C,D} to the desired {A',B',C',D'} reference position,
considering the vehicle grille and the license plate in the same plan.
Results
All ours experiments have been carried out on the Training Base and on the Test Base.
The former is used to produce the oriented-contour point models of the vehicle classes.
While the last is utilized to evaluate the performance of the classification system.
The last results of my algorithm are publied in my
PhD thesis:
94.2 % of well classifications.
In page 183 you can find the confusion matrix.
Using the Databases
The databases, whose copyrights belong to the
Institut des Systèmes Intelligents et Robotique
and
LPREditor, are free of charge.
Any commercial use of the databases is strictly forbidden.
All the vehicles' license plates where blurred to avoid violation of privacy rights.
Please give appropriate acknowledgements when using these databases and quote
the following paper :
[ICPR06]
Pablo Negri, Xavier Clady, Maurice Milgram, Raphael Poulenard,
"An Oriented-Contour Point Based Voting Algorithm for Vehicle Type Classification"
, International Conference of Pattern Recognition, Hong Kong, 20-24 august 2006
[
bib].
These databases can be referred as "ISIR-LPReditor Make & Model Vehicle
Recognition DataBases" (ISIR_LPReditor MMVR).
Should you wish to use the databases, please send us an email with a brief
explanation of the intended usage (in English, French or Spanish) and we will
immediately provide the passwords to unzip the files:
pnegri [ a ] telecentro . com . ar
xavier . clady [ a ] upmc . fr
Download