Method of vehicle identification and a system for vehicle identification

• US 9,396,403 B2
• Filed: 12/31/2012
• Issued: 07/19/2016
• Est. Priority Date: 12/31/2012
• Status: Expired due to Fees

• Current Assignee: Instytut Badawczy Drog I Mostow
• Original Assignee: Instytut Badawczy Drog I Mostow
• Inventors: Dzienis, Tadeusz, Karkowski, Michal, Dolega, Cezary, Mrowka, Pawel
• Primary Examiner(s): Desire, Gregory M
• Application Number:
  US14/359,306
• Publication Number:
  US 20150294174A1
• Time in Patent Office: 1,296 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/5838 using colour
• CPC Class Codes: G06T 1/0007 Image acquisition
• CPC Class Codes: G06T 2207/10004 Still image; Photographic image
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/10024 Color image
• CPC Class Codes: G06T 2207/30252 Vehicle exterior; Vicinity of vehicle
• CPC Class Codes: G06T 3/40 Scaling of whole images or parts thereof, e.g. expanding or contracting
• CPC Class Codes: G06T 3/60 Rotation of whole images or parts thereof
• CPC Class Codes: G06V 20/62 Text, e.g. of license plates, overlay texts or captions on TV images
• CPC Class Codes: G06V 20/625 License plates
• CPC Class Codes: G06V 30/224 of printed characters having additional code marks or containing code marks

Abstract

A method for vehicle identification to determine at least one vehicle characteristic, comprising: obtaining an input image (301) of the vehicle from an image source (101); normalization of the input image (301) of the vehicle in a normalization unit (103, 104) to obtain a normalized image; determining the vehicle characteristic in a classification unit (111, 112) by comparing parameters of a normalized image obtained in a parametrization unit (107, 108) with parameters of reference images obtained from a reference database (113, 114). Normalization in the normalization unit (103, 104) comprises the steps of: detecting a registration plate area (303) within the input image (301); processing the input image (301) basing on normalization attributes defining at least one scaling coefficient (z_norm, a_norm, v_norm); choosing from the scaled image (307) a RoI area (308) of a normalized size and location dependent on the location of the registration plate area (303); and presenting data from the RoI area (308) as a normalized image (311).

Claims

1. A computer implemented method for vehicle type identification to determine at least one vehicle characteristic, the method comprising:

obtaining an input image of the vehicle from an image source connected to a motion sensor and configured to be triggered the motion sensor;

converting the input image of the vehicle by a processor of a normalization unit to a normalized image by;

reading normalization attributes (z_norm, a_norm, v_norm) from a unit for calibration of normalization attributes;

detecting a registration plate area within the input image processing the input image basing on normalization attributes defining at least one scaling coefficient (z_norm, a_norm, v_norm);

choosing from the sealed image a RoI area of a normalized size and location dependent on the location of the registration plate area; and

, presenting data from the RoI area as a normalized image;

performing by a processor of a parameterization unit a smoothing filtration of the normalized image to obtain a smoothed image (I_i);

determining oriented gradient images (O_i,j) of brightness change for the smoothed image (h) in different directions (j);

dividing the oriented gradient images (O_i,j) into blocks (B_i,j,n);

determining a column vector (v_i,n) having R dimensions and defined as

v_i,n=[∇

(B_i,0,n), . . . ,∇

(B_i,j,n)]

reducing dimension of the column vector (v_i,n) by left multiplication by a casting matrix P_i,n having S dimensions, wherein S is lower than R to obtain a column vector c^(I)_i,n defined as;

c^(I)_i,n=P_i,n*v_i,n

providing the column vector c as the output of the parameterisation unit describing parameters of the normalized image;

and determining the vehicle characteristic by a processor of a classification unit by comparing the parameters of a normalized image with parameters of reference images obtained from a reference database.

2. The method according to claim 1, wherein for a specific input image scene the following steps are performed:

obtaining a set of reference images;

selecting values of normalization attributes defining at least one scale coefficient (z_norm, a_norm, v_norm) in a manner allowing to obtain the scaled image with the registration plate of the normalized size (w_plate, h_plate) as a result of processing during normalization, for most of the reference images;

setting said selected values of normalization attributes (z_norm, a_norm, v_norm) to perform normalization of subsequent input images obtained in given scene.

3. The method according to claim 2, further comprising selecting a value of a normalization attribute determining angular rotation of image (α

_norm), in a manner allowing to obtain a rotated image with a registration plate of a lower border positioned at a normalized angle as a result of processing during normalization, for most of the reference images.

4. The method according to claim 1, wherein processing of input image during the normalization results in color reduction of input image.

5. The method according to claim 1, wherein processing of input image during the normalization results in decimation of input image.

6. The method according to claim 1, further comprising performing contrast adjustment within the RoI area, presented as the normalized image.

7. The method according to claim 1, further comprising decimating the smoothed image (I0) to obtain a second smoothed image (I1) of a reduced scale and subjecting the first and second images (I0, I1) to further parameterization independently.

8. The method according to claim 1, wherein:

within the RoI area, presented as the normalized image, the field comprising the registration plate and/or the field comprising the area from beyond input image are presented as undefined data

blocks (Bi,j,n) of the oriented gradient images (Oi,j) in which at least one pixel has been described as containing undefined data are excluded from analysis (605-607) in the normalization process.

9. The method according to claim 1, further comprising subjecting each element(s) of the column vector c^(I)_i,n to nonlinear transformation to obtain a transformed column vector comprising elements c^(nI)_i,n,s according to a formula:

c_i,n,s^(nI)=tan h(a_i,n,s·

(c_i,n,s^(I)−

b_i,n,s))

wherein values of parameters a_i,n,s and n_i,n,s are determined in a learning process and providing the transformed vector c comprising elements c^(nI)_i,n,s as the output of the parameterisation unit describing parameters of the normalized image.

10. A system for vehicle type identification to determine at least one vehicle characteristic, the system comprising:

an image source providing an input image of the vehicle, the image source being connected to a motion sensor and configured to capture an image when the image source is triggered by the motion sensor;

a normalization unit having a processor configured to normalize of the input image of the vehicle to obtain a normalized image by;

reading normalization attributes (z_norm, a_norm, v_norm) from a unit for calibration of normalization attributes;

detecting a registration plate area within the input image

processing the input image basing on normalization attributes defining at least one scaling coefficient (z_norm, a_norm, v_norm);

choosing from the scaled image a RoI area of a normalized size and location dependent on the location of the registration plate area;

and presenting data from the RoI area as a normalized image;

and a classification unit having processor configured to determine the vehicle characteristic by comparing parameters of a normalized image obtained by a processor of a parameterization unit with parameters of reference images obtained from a reference database;

wherein the processor of the parameterization unit is further configured to;

perform a smoothing filtration of the normalized image to obtain a smoothed image (Ii);

determine oriented gradient images (Oi,j) of brightness change for the smoothed image (Ii) in different directions (i);

divide the oriented gradient images (Oi,j) into blocks (Bi,j,n);

determine a column vector (v_i,n) having R dimensions and defined as

v_i,n=[∇

(B_i,0,n), . . . , ∇

(B_i,j,n)]

reduce dimension of the column vector (vin) by left multiplication by a casting matrix P_i,n, having S dimensions, wherein S is lower than R to obtain a column vector c^(I)_i,n defined as

c^(I)_i,n=P_i,n*v_i,n

provide the column vector c as the output describing parameters of the normalized image.